Mobile Ad Hoc Networks:
Routing, MAC and Transport Issues
Nitin H. Vaidya
University of Illinois at Urbana-Champaign
nhv@uiuc.edu
http://www.crhc.uiuc.edu/~nhv
© 2004 Nitin Vaidya
1
Notes

Names in brackets, as in [Xyz00], refer to a
document in the list of references

The handout may not be as readable as the original
slides, since the slides contain colored text and
figures
Note that different colors in the colored slides may look
identically black in the black-and-white handout
2
Statutory Warnings

Only most important features of various schemes are typically
discussed, i.e, features I consider as being important
Others may disagree

Most schemes include many more details, and optimizations
Not possible to cover all details in this tutorial

Be aware that some protocol specs have changed several
times, and the slides may not reflect the most current
specifications

Jargon used to discuss a scheme may occasionally differ from
that used by the proposers
3
Coverage

Not intended to be exhaustive

Many interesting papers not covered in the tutorial
due to lack of time
4
Tutorial Outline











Introduction
Unicast routing
Multicast routing
Geocast routing
Medium Access Control
Performance of UDP and TCP
Security Issues
Implementation Issues
Distributed Algorithms
Standards activities
Open problems
5
Mobile Ad Hoc Networks (MANET)
Introduction and Generalities
6
Mobile Ad Hoc Networks

Formed by wireless hosts which may be mobile

Without (necessarily) using a pre-existing
infrastructure

Routes between nodes may potentially contain
multiple hops
7
Mobile Ad Hoc Networks

May need to traverse multiple links to reach a
destination
8
Mobile Ad Hoc Networks (MANET)

Mobility causes route changes
9
Why Ad Hoc Networks ?

Ease of deployment

Speed of deployment

Decreased dependence on infrastructure
10
Many Applications




Personal area networking
cell phone, laptop, ear phone, wrist watch
Military environments
soldiers, tanks, planes
Civilian environments
taxi cab network
meeting rooms
sports stadiums
boats, small aircraft
Emergency operations
search-and-rescue
policing and fire fighting
11
Many Variations

Fully Symmetric Environment
all nodes have identical capabilities and responsibilities

Asymmetric Capabilities
transmission ranges and radios may differ
battery life at different nodes may differ
processing capacity may be different at different nodes
speed of movement

Asymmetric Responsibilities
only some nodes may route packets
some nodes may act as leaders of nearby nodes (e.g.,
cluster head)
12
Many Variations

Traffic characteristics may differ in different ad hoc
networks
bit rate
timeliness constraints
reliability requirements
unicast / multicast / geocast
host-based addressing / content-based addressing /
capability-based addressing

May co-exist (and co-operate) with an infrastructurebased network
13
Many Variations

Mobility patterns may be different
people sitting at an airport lounge
New York taxi cabs
kids playing
military movements
personal area network

Mobility characteristics
speed
predictability
• direction of movement
• pattern of movement
uniformity (or lack thereof) of mobility characteristics among
different nodes
14
Challenges








Limited wireless transmission range
Broadcast nature of the wireless medium
Hidden terminal problem (see next slide)
Packet losses due to transmission errors
Mobility-induced route changes
Mobility-induced packet losses
Battery constraints
Potentially frequent network partitions
Ease of snooping on wireless transmissions (security
hazard)
15
Hidden Terminal Problem
A
B
C
Nodes A and C cannot hear each other
Transmissions by nodes A and C can collide at node B
Nodes A and C are hidden from each other
16
Research on Mobile Ad Hoc Networks
Variations in capabilities & responsibilities
X
Variations in traffic characteristics, mobility models, etc.
X
Performance criteria (e.g., optimize throughput, reduce
energy consumption)
+
Increased research funding
=
Significant research activity
17
The Holy Grail

A one-size-fits-all solution
Perhaps using an adaptive/hybrid approach that can adapt
to situation at hand

Difficult problem

Many solutions proposed trying to address a
sub-space of the problem domain
18
Assumption

Unless stated otherwise, fully symmetric environment
is assumed implicitly
all nodes have identical capabilities and responsibilities
19
Unicast Routing
in
Mobile Ad Hoc Networks
20
Why is Routing in MANET different ?

Host mobility
link failure/repair due to mobility may have different
characteristics than those due to other causes

Rate of link failure/repair may be high when nodes
move fast

New performance criteria may be used
route stability despite mobility
energy consumption
21
Unicast Routing Protocols

Many protocols have been proposed

Some have been invented specifically for MANET

Others are adapted from previously proposed
protocols for wired networks

No single protocol works well in all environments
some attempts made to develop adaptive protocols
22
Routing Protocols

Proactive protocols
Determine routes independent of traffic pattern
Traditional link-state and distance-vector routing protocols
are proactive

Reactive protocols
Maintain routes only if needed

Hybrid protocols
23
Trade-Off

Latency of route discovery
Proactive protocols may have lower latency since routes are
maintained at all times
Reactive protocols may have higher latency because a route
from X to Y will be found only when X attempts to send to Y

Overhead of route discovery/maintenance
Reactive protocols may have lower overhead since routes
are determined only if needed
Proactive protocols can (but not necessarily) result in higher
overhead due to continuous route updating

Which approach achieves a better trade-off depends
24
on the traffic and mobility patterns
Overview of Unicast Routing Protocols
25
Flooding for Data Delivery

Sender S broadcasts data packet P to all its
neighbors

Each node receiving P forwards P to its neighbors

Sequence numbers used to avoid the possibility of
forwarding the same packet more than once

Packet P reaches destination D provided that D is
reachable from sender S

Node D does not forward the packet
26
Flooding for Data Delivery
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
N
Represents a node that has received packet P
Represents that connected nodes are within each
other’s transmission range
27
Flooding for Data Delivery
Y
Broadcast transmission
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
N
Represents a node that receives packet P for
the first time
Represents transmission of packet P
28
Flooding for Data Delivery
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
N
• Node H receives packet P from two neighbors:
potential for collision
29
Flooding for Data Delivery
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
• Node C receives packet P from G and H, but does not forward
it again, because node C has already forwarded packet P once
30
Flooding for Data Delivery
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
• Nodes J and K both broadcast packet P to node D
• Since nodes J and K are hidden from each other, their
transmissions may collide
=> Packet P may not be delivered to node D at all,
despite the use of flooding
31
Flooding for Data Delivery
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
N
• Node D does not forward packet P, because node D
is the intended destination of packet P
32
Flooding for Data Delivery
Y
Z
S
E
F
B
C
M
J
A
L
G
H
• Flooding completed
K
I
D
N
• Nodes unreachable from S do not receive packet P (e.g., node Z)
• Nodes for which all paths from S go through the destination D
also do not receive packet P (example: node N)
33
Flooding for Data Delivery
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
• Flooding may deliver packets to too many nodes
(in the worst case, all nodes reachable from sender
may receive the packet)
D
N
34
Flooding for Data Delivery: Advantages

Simplicity

May be more efficient than other protocols when rate
of information transmission is low enough that the
overhead of explicit route discovery/maintenance
incurred by other protocols is relatively higher
this scenario may occur, for instance, when nodes transmit
small data packets relatively infrequently, and many topology
changes occur between consecutive packet transmissions

Potentially higher reliability of data delivery
Because packets may be delivered to the destination on
multiple paths
35
Flooding for Data Delivery: Disadvantages

Potentially, very high overhead
Data packets may be delivered to too many nodes who do
not need to receive them

Potentially lower reliability of data delivery
Flooding uses broadcasting -- hard to implement reliable
broadcast delivery without significantly increasing overhead
– Broadcasting in IEEE 802.11 MAC is unreliable
In our example, nodes J and K may transmit to node D
simultaneously, resulting in loss of the packet
– in this case, destination would not receive the packet at all
36
Flooding of Control Packets

Many protocols perform (potentially limited) flooding
of control packets, instead of data packets

The control packets are used to discover routes

Discovered routes are subsequently used to send
data packet(s)

Overhead of control packet flooding is amortized over
data packets transmitted between consecutive
control packet floods
37
Dynamic Source Routing (DSR) [Johnson96]

When node S wants to send a packet to node D, but
does not know a route to D, node S initiates a route
discovery

Source node S floods Route Request (RREQ)

Each node appends own identifier when forwarding
RREQ
38
Route Discovery in DSR
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
Represents a node that has received RREQ for D from S
39
Route Discovery in DSR
Y
Broadcast transmission
[S]
S
Z
E
F
B
C
M
J
A
L
G
H
K
I
D
N
Represents transmission of RREQ
[X,Y]
Represents list of identifiers appended to RREQ
40
Route Discovery in DSR
Y
Z
S
E
[S,E]
F
B
C
A
M
J
[S,C]
H
G
K
I
L
D
N
• Node H receives packet RREQ from two neighbors:
potential for collision
41
Route Discovery in DSR
Y
Z
S
E
F
B
[S,E,F]
C
M
J
A
L
G
H
I
[S,C,G] K
D
N
• Node C receives RREQ from G and H, but does not forward
it again, because node C has already forwarded RREQ once
42
Route Discovery in DSR
Y
Z
S
E
[S,E,F,J]
F
B
C
M
J
A
L
G
H
K
I
D
[S,C,G,K]
• Nodes J and K both broadcast RREQ to node D
• Since nodes J and K are hidden from each other, their
transmissions may collide
N
43
Route Discovery in DSR
Y
Z
S
E
[S,E,F,J,M]
F
B
C
M
J
A
L
G
H
K
D
I
• Node D does not forward RREQ, because node D
is the intended target of the route discovery
N
44
Route Discovery in DSR

Destination D on receiving the first RREQ, sends a
Route Reply (RREP)

RREP is sent on a route obtained by reversing the
route appended to received RREQ

RREP includes the route from S to D on which RREQ
was received by node D
45
Route Reply in DSR
Y
Z
S
E
RREP [S,E,F,J,D]
F
B
C
M
J
A
L
G
H
K
I
Represents RREP control message
D
N
46
Route Reply in DSR

Route Reply can be sent by reversing the route in
Route Request (RREQ) only if links are guaranteed
to be bi-directional
To ensure this, RREQ should be forwarded only if it received
on a link that is known to be bi-directional

If unidirectional (asymmetric) links are allowed, then
RREP may need a route discovery for S from node D
Unless node D already knows a route to node S
If a route discovery is initiated by D for a route to S, then the
Route Reply is piggybacked on the Route Request from D.

If IEEE 802.11 MAC is used to send data, then links
have to be bi-directional (since Ack is used)
47
Dynamic Source Routing (DSR)

Node S on receiving RREP, caches the route
included in the RREP

When node S sends a data packet to D, the entire
route is included in the packet header
hence the name source routing

Intermediate nodes use the source route included in
a packet to determine to whom a packet should be
forwarded
48
Data Delivery in DSR
Y
DATA [S,E,F,J,D]
S
Z
E
F
B
C
M
J
A
L
G
H
K
I
D
N
Packet header size grows with route length
49
When to Perform a Route Discovery

When node S wants to send data to node D, but does
not know a valid route node D
50
DSR Optimization: Route Caching






Each node caches a new route it learns by any
means
When node S finds route [S,E,F,J,D] to node D, node
S also learns route [S,E,F] to node F
When node K receives Route Request [S,C,G]
destined for node, node K learns route [K,G,C,S] to
node S
When node F forwards Route Reply RREP
[S,E,F,J,D], node F learns route [F,J,D] to node D
When node E forwards Data [S,E,F,J,D] it learns
route [E,F,J,D] to node D
A node may also learn a route when it overhears
51
Data packets
Use of Route Caching

When node S learns that a route to node D is broken,
it uses another route from its local cache, if such a
route to D exists in its cache. Otherwise, node S
initiates route discovery by sending a route request

Node X on receiving a Route Request for some node
D can send a Route Reply if node X knows a route to
node D

Use of route cache
can speed up route discovery
can reduce propagation of route requests
52
Use of Route Caching
[S,E,F,J,D]
[E,F,J,D]
S
[F,J,D],[F,E,S]
E
F
B
[J,F,E,S]
C
J
[C,S]
A
M
L
G
H
[G,C,S]
D
K
I
N
Z
[P,Q,R] Represents cached route at a node
(DSR maintains the cached routes in a tree format)
53
Use of Route Caching:
Can Speed up Route Discovery
[S,E,F,J,D]
[E,F,J,D]
S
[F,J,D],[F,E,S]
E
F
B
C
[G,C,S]
[C,S]
A
[J,F,E,S]
M
J
L
G
H
I
[K,G,C,S] K
D
RREP
N
RREQ
When node Z sends a route request
for node C, node K sends back a route
reply [Z,K,G,C] to node Z using a locally
cached route
Z
54
Use of Route Caching:
Can Reduce Propagation of Route Requests
[S,E,F,J,D]
Y
[E,F,J,D]
S
[F,J,D],[F,E,S]
E
F
B
C
[G,C,S]
[C,S]
A
[J,F,E,S]
M
J
L
G
H
I
D
[K,G,C,S] K
RREP
N
RREQ
Z
Assume that there is no link between D and Z.
Route Reply (RREP) from node K limits flooding of RREQ.
In general, the reduction may be less dramatic.
55
Route Error (RERR)
Y
RERR [J-D]
S
Z
E
F
B
C
M
J
A
L
G
H
K
I
D
N
J sends a route error to S along route J-F-E-S when its attempt to
forward the data packet S (with route SEFJD) on J-D fails
Nodes hearing RERR update their route cache to remove link J-D
56
Route Caching: Beware!

Stale caches can adversely affect performance

With passage of time and host mobility, cached
routes may become invalid

A sender host may try several stale routes (obtained
from local cache, or replied from cache by other
nodes), before finding a good route

An illustration of the adverse impact on TCP will be
discussed later in the tutorial [Holland99]
57
Dynamic Source Routing: Advantages

Routes maintained only between nodes who need to
communicate
reduces overhead of route maintenance

Route caching can further reduce route discovery
overhead

A single route discovery may yield many routes to the
destination, due to intermediate nodes replying from
local caches
58
Dynamic Source Routing: Disadvantages

Packet header size grows with route length due to
source routing

Flood of route requests may potentially reach all
nodes in the network

Care must be taken to avoid collisions between route
requests propagated by neighboring nodes
insertion of random delays before forwarding RREQ

Increased contention if too many route replies come
back due to nodes replying using their local cache
Route Reply Storm problem
Reply storm may be eased by preventing a node from
sending RREP if it hears another RREP with a shorter route
59
Dynamic Source Routing: Disadvantages

An intermediate node may send Route Reply using a
stale cached route, thus polluting other caches

This problem can be eased if some mechanism to
purge (potentially) invalid cached routes is
incorporated.

For some proposals for cache invalidation, see
[Hu00Mobicom]
Static timeouts
Adaptive timeouts based on link stability
60
Flooding of Control Packets

How to reduce the scope of the route request flood ?
LAR [Ko98Mobicom]
Query localization [Castaneda99Mobicom]

How to reduce redundant broadcasts ?
The Broadcast Storm Problem [Ni99Mobicom]
61
Location-Aided Routing (LAR) [Ko98Mobicom]

Exploits location information to limit scope of route
request flood
Location information may be obtained using GPS

Expected Zone is determined as a region that is
expected to hold the current location of the
destination
Expected region determined based on potentially old
location information, and knowledge of the destination’s
speed

Route requests limited to a Request Zone that
contains the Expected Zone and location of the
sender node
62
Expected Zone in LAR
X = last known location of node
D, at time t0
Y = location of node D at current
time t1, unknown to node S
r = (t1 - t0) * estimate of D’s speed
r
X
Y
Expected Zone
63
Request Zone in LAR
Network Space
Request Zone
r
B
A
X
Y
S
64
LAR

Only nodes within the request zone forward route
requests
Node A does not forward RREQ, but node B does (see
previous slide)

Request zone explicitly specified in the route request

Each node must know its physical location to
determine whether it is within the request zone
65
LAR

Only nodes within the request zone forward route
requests

If route discovery using the smaller request zone fails
to find a route, the sender initiates another route
discovery (after a timeout) using a larger request
zone
the larger request zone may be the entire network

Rest of route discovery protocol similar to DSR
66
LAR Variations: Adaptive Request Zone

Each node may modify the request zone included in
the forwarded request

Modified request zone may be determined using
more recent/accurate information, and may be
smaller than the original request zone
B
S
Request zone adapted by B
Request zone defined by sender S
67
LAR Variations: Implicit Request Zone

In the previous scheme, a route request explicitly
specified a request zone

Alternative approach: A node X forwards a route
request received from Y if node X is deemed to be
closer to the expected zone as compared to Y

The motivation is to attempt to bring the route request
physically closer to the destination node after each
forwarding
68
Location-Aided Routing


The basic proposal assumes that, initially, location
information for node X becomes known to Y only
during a route discovery
This location information is used for a future route
discovery
Each route discovery yields more updated information which
is used for the next discovery
Variations
 Location information can also be piggybacked on any
message from Y to X
 Y may also proactively distribute its location
information
Similar to other protocols discussed later (e.g., DREAM, 69
GLS)
Location Aided Routing (LAR)

Advantages
reduces the scope of route request flood
reduces overhead of route discovery

Disadvantages
Nodes need to know their physical locations
Does not take into account possible existence of
obstructions for radio transmissions
70
Detour
Routing Using Location Information
71
Geographic Distance Routing (GEDIR) [Lin98]




Location of the destination node is assumed known
Each node knows location of its neighbors
Each node forwards a packet to its neighbor closest
to the destination
Route taken from S to D shown below
H
A
S
D
B
E
F
C
G
obstruction
72
Geographic Distance Routing (GEDIR)
[Stojmenovic99]

The algorithm terminates when same edge traversed
twice consecutively

Algorithm fails to route from S to E
Node G is the neighbor of C who is closest from destination
E, but C does not have a route to E
H
A
S
D
B
E
F
C
G
obstruction
73
Routing with Guaranteed Delivery
[Bose99Dialm]

Improves on GEDIR [Lin98]

Guarantees delivery (using location information)
provided that a path exists from source to destination

Routes around obstacles if necessary

A similar idea also appears in [Karp00Mobicom]
74
End of
Detour
Back to
Reducing Scope of
the Route Request Flood
75
Query Localization [Castaneda99Mobicom]

Limits route request flood without using physical
information

Route requests are propagated only along paths that
are close to the previously known route

The closeness property is defined without using
physical location information
76
Query Localization

Path locality heuristic: Look for a new path that
contains at most k nodes that were not present in the
previously known route

Old route is piggybacked on a Route Request

Route Request is forwarded only if the accumulated
route in the Route Request contains at most k new
nodes that were absent in the old route
this limits propagation of the route request
77
Query Localization: Example
G
G
F
F
E
Node D moved
B
C
A
D
Node F does not forward the route
request since it is not on any route
from S to D that contains at most
2 new nodes
E
D
B
C
A
Permitted routes
with k = 2
Initial route
from S to D
S
S
78
Query Localization

Advantages:
Reduces overhead of route discovery without using physical
location information
Can perform better in presence of obstructions by searching
for new routes in the vicinity of old routes

Disadvantage:
May yield routes longer than LAR
(Shortest route may contain more than k new nodes)
79
Broadcast Storm Problem [Ni99Mobicom]




When node A broadcasts a route query, nodes B and
C both receive it
B and C both forward to their neighbors
B and C transmit at about the same time since they
are reacting to receipt of the same message from A
This results in a high probability of collisions
D
B
C
A
80
Broadcast Storm Problem


Redundancy: A given node may receive the same
route request from too many nodes, when one copy
would have sufficed
Node D may receive from nodes B and C both
D
B
C
A
81
Solutions for Broadcast Storm

Probabilistic scheme: On receiving a route request for
the first time, a node will re-broadcast (forward) the
request with probability p

Also, re-broadcasts by different nodes should be
staggered by using a collision avoidance technique (wait
a random delay when channel is idle)
this would reduce the probability that nodes B and C would
forward a packet simultaneously in the previous example
82
Solutions for Broadcast Storms


Counter-Based Scheme: If node E hears more than k
neighbors broadcasting a given route request, before
it can itself forward it, then node E will not forward the
request
Intuition: k neighbors together have probably already
forwarded the request to all of E’s neighbors
D
E
B
C
F
A
83
Solutions for Broadcast Storms

Distance-Based Scheme: If node E hears RREQ
broadcasted by some node Z within physical distance
d, then E will not re-broadcast the request

Intuition: Z and E are too close, so transmission
areas covered by Z and E are not very different
 if E re-broadcasts the request, not many nodes who have not
already heard the request from Z will hear the request
E
<d
Z
84
Summary: Broadcast Storm Problem

Flooding is used in many protocols, such as Dynamic
Source Routing (DSR)

Problems associated with flooding
collisions
redundancy

Collisions may be reduced by “jittering” (waiting for a
random interval before propagating the flood)

Redundancy may be reduced by selectively rebroadcasting packets from only a subset of the nodes
85
Ad Hoc On-Demand Distance Vector Routing
(AODV) [Perkins99Wmcsa]

DSR includes source routes in packet headers

Resulting large headers can sometimes degrade
performance
particularly when data contents of a packet are small

AODV attempts to improve on DSR by maintaining
routing tables at the nodes, so that data packets do
not have to contain routes

AODV retains the desirable feature of DSR that
routes are maintained only between nodes which
need to communicate
86
AODV

Route Requests (RREQ) are forwarded in a manner
similar to DSR

When a node re-broadcasts a Route Request, it sets
up a reverse path pointing towards the source
AODV assumes symmetric (bi-directional) links

When the intended destination receives a Route
Request, it replies by sending a Route Reply

Route Reply travels along the reverse path set-up
when Route Request is forwarded
87
Route Requests in AODV
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
Represents a node that has received RREQ for D from S
88
Route Requests in AODV
Y
Broadcast transmission
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
Represents transmission of RREQ
89
Route Requests in AODV
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
N
Represents links on Reverse Path
90
Reverse Path Setup in AODV
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
• Node C receives RREQ from G and H, but does not forward
it again, because node C has already forwarded RREQ once
91
Reverse Path Setup in AODV
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
92
Reverse Path Setup in AODV
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
• Node D does not forward RREQ, because node D
is the intended target of the RREQ
N
93
Route Reply in AODV
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
N
Represents links on path taken by RREP
94
Route Reply in AODV

An intermediate node (not the destination) may also
send a Route Reply (RREP) provided that it knows a
more recent path than the one previously known to
sender S

To determine whether the path known to an
intermediate node is more recent, destination
sequence numbers are used

The likelihood that an intermediate node will send a
Route Reply when using AODV not as high as DSR
A new Route Request by node S for a destination is
assigned a higher destination sequence number. An
intermediate node which knows a route, but with a smaller
sequence number, cannot send Route Reply
95
Forward Path Setup in AODV
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
N
Forward links are setup when RREP travels along
the reverse path
Represents a link on the forward path
96
Data Delivery in AODV
Y
DATA
Z
S
E
F
B
C
M
J
A
L
G
H
K
D
I
N
Routing table entries used to forward data packet.
Route is not included in packet header.
97
Timeouts

A routing table entry maintaining a reverse path is
purged after a timeout interval
timeout should be long enough to allow RREP to come back

A routing table entry maintaining a forward path is
purged if not used for a active_route_timeout interval
if no is data being sent using a particular routing table entry,
that entry will be deleted from the routing table (even if the
route may actually still be valid)
98
Link Failure Reporting

A neighbor of node X is considered active for a
routing table entry if the neighbor sent a packet within
active_route_timeout interval which was forwarded
using that entry

When the next hop link in a routing table entry
breaks, all active neighbors are informed

Link failures are propagated by means of Route Error
messages, which also update destination sequence
numbers
99
Route Error

When node X is unable to forward packet P (from
node S to node D) on link (X,Y), it generates a RERR
message

Node X increments the destination sequence number
for D cached at node X

The incremented sequence number N is included in
the RERR

When node S receives the RERR, it initiates a new
route discovery for D using destination sequence
number at least as large as N
100
Destination Sequence Number

Continuing from the previous slide …

When node D receives the route request with
destination sequence number N, node D will set its
sequence number to N, unless it is already larger
than N
101
Link Failure Detection

Hello messages: Neighboring nodes periodically
exchange hello message

Absence of hello message is used as an indication of
link failure

Alternatively, failure to receive several MAC-level
acknowledgement may be used as an indication of
link failure
102
Why Sequence Numbers in AODV

To avoid using old/broken routes
To determine which route is newer

To prevent formation of loops
A
B
C
D
E
Assume that A does not know about failure of link C-D
because RERR sent by C is lost
Now C performs a route discovery for D. Node A receives
the RREQ (say, via path C-E-A)
Node A will reply since A knows a route to D via node B
Results in a loop (for instance, C-E-A-B-C )
103
Why Sequence Numbers in AODV
A
B
C
D
E
Loop C-E-A-B-C
104
Optimization: Expanding Ring Search

Route Requests are initially sent with small Time-toLive (TTL) field, to limit their propagation
DSR also includes a similar optimization

If no Route Reply is received, then larger TTL tried
105
Summary: AODV

Routes need not be included in packet headers

Nodes maintain routing tables containing entries only
for routes that are in active use

At most one next-hop per destination maintained at
each node
DSR may maintain several routes for a single destination

Unused routes expire even if topology does not
change
106
So far ...

All protocols discussed so far perform some form of
flooding

Now we will consider protocols which try to
reduce/avoid such behavior
107
Link Reversal Algorithm [Gafni81]
A
B
F
C
E
G
D
108
Link Reversal Algorithm
A
B
F
Links are bi-directional
But algorithm imposes
logical directions on them
C
E
D
G
Maintain a directed acyclic
graph (DAG) for each
destination, with the destination
being the only sink
This DAG is for destination
node D
109
Link Reversal Algorithm
A
B
F
C
E
G
Link (G,D) broke
D
Any node, other than the destination, that has no outgoing links
reverses all its incoming links.
110
Node G has no outgoing links
Link Reversal Algorithm
A
B
F
C
E
G
Represents a
link that was
reversed recently
D
Now nodes E and F have no outgoing links
111
Link Reversal Algorithm
A
B
F
C
E
G
Represents a
link that was
reversed recently
D
Now nodes B and G have no outgoing links
112
Link Reversal Algorithm
A
B
F
C
E
G
Represents a
link that was
reversed recently
D
Now nodes A and F have no outgoing links
113
Link Reversal Algorithm
A
B
F
C
E
G
Represents a
link that was
reversed recently
D
Now all nodes (other than destination D) have an outgoing link
114
Link Reversal Algorithm
A
B
F
C
E
G
D
DAG has been restored with only the destination as a sink
115
Link Reversal Algorithm

Attempts to keep link reversals local to where the
failure occurred
But this is not guaranteed

When the first packet is sent to a destination, the
destination oriented DAG is constructed

The initial construction does result in flooding of
control packets
116
Link Reversal Algorithm

The previous algorithm is called a full reversal
method since when a node reverses links, it reverses
all its incoming links

Partial reversal method [Gafni81]: A node reverses
incoming links from only those neighbors who have
not themselves reversed links “previously”
If all neighbors have reversed links, then the node reverses
all its incoming links
“Previously” at node X means since the last link reversal
done by node X
117
Partial Reversal Method
A
B
F
C
E
G
Link (G,D) broke
D
Node G has no outgoing links
118
Partial Reversal Method
A
B
F
C
E
G
D
Represents a
link that was
reversed recently
Represents a
node that has
reversed links
Now nodes E and F have no outgoing links
119
Partial Reversal Method
A
B
F
C
E
G
Represents a
link that was
reversed recently
D
Nodes E and F do not reverse links from node G
Now node B has no outgoing links
120
Partial Reversal Method
A
B
F
C
E
G
Represents a
link that was
reversed recently
D
Now node A has no outgoing links
121
Partial Reversal Method
A
B
F
C
E
G
Represents a
link that was
reversed recently
D
Now all nodes (except destination D) have outgoing links
122
Partial Reversal Method
A
B
F
C
E
G
D
DAG has been restored with only the destination as a sink
123
Link Reversal Methods: Advantages

Link reversal methods attempt to limit updates to
routing tables at nodes in the vicinity of a broken link
Partial reversal method tends to be better than full reversal
method

Each node may potentially have multiple routes to a
destination
124
Link Reversal Methods: Disadvantage

Need a mechanism to detect link failure
hello messages may be used
but hello messages can add to contention

If network is partitioned, link reversals continue
indefinitely
125
Link Reversal in a Partitioned Network
A
B
F
C
E
G
D
This DAG is for destination node D
126
Full Reversal in a Partitioned Network
A
B
F
C
E
G
D
A and G do not have outgoing links
127
Full Reversal in a Partitioned Network
A
B
F
C
E
G
D
E and F do not have outgoing links
128
Full Reversal in a Partitioned Network
A
B
F
C
E
G
D
B and G do not have outgoing links
129
Full Reversal in a Partitioned Network
A
B
F
C
E
G
D
E and F do not have outgoing links
130
Full Reversal in a Partitioned Network
A
B
F
C
E
G
In the partition
disconnected from
destination D, link
reversals continue, until
the partitions merge
Need a mechanism to
minimize this wasteful
activity
D
Similar scenario can
occur with partial
reversal method too
131
Temporally-Ordered Routing Algorithm
(TORA) [Park97Infocom]

TORA modifies the partial link reversal method to be
able to detect partitions

When a partition is detected, all nodes in the partition
are informed, and link reversals in that partition cease
132
Partition Detection in TORA
B
DAG for
destination D
A
C
E
D
F
133
Partition Detection in TORA
B
A
C
E
D
F
Node A has no outgoing links
TORA uses a
modified partial
reversal method
134
Partition Detection in TORA
B
A
C
E
D
F
Node B has no outgoing links
TORA uses a
modified partial
reversal method
135
Partition Detection in TORA
B
A
C
E
D
F
Node B has no outgoing links
136
Partition Detection in TORA
B
A
C
E
D
F
Node C has no outgoing links -- all its neighbor have
reversed links previously.
137
Partition Detection in TORA
B
A
C
E
D
F
Nodes A and B receive the reflection from node C
Node B now has no outgoing link
138
Partition Detection in TORA
B
A
C
E
Node B propagates
the reflection to node A
D
F
Node A has received the reflection from all its neighbors.
Node A determines that it is partitioned from destination D.
139
Partition Detection in TORA
B
A
C
E
On detecting a partition,
node A sends a clear (CLR)
message that purges all
directed links in that
partition
D
F
140
TORA

Improves on the partial link reversal method in
[Gafni81] by detecting partitions and stopping nonproductive link reversals

Paths may not be shortest

The DAG provides many hosts the ability to send
packets to a given destination
Beneficial when many hosts want to communicate with a
single destination
141
TORA Design Decision



TORA performs link reversals as dictated by
[Gafni81]
However, when a link breaks, it looses its direction
When a link is repaired, it may not be assigned a
direction, unless some node has performed a route
discovery after the link broke
if no one wants to send packets to D anymore, eventually,
the DAG for destination D may disappear

TORA makes effort to maintain the DAG for D only if
someone needs route to D
142
Reactive behavior
TORA Design Decision

One proposal for modifying TORA optionally allowed
a more proactive behavior, such that a DAG would be
maintained even if no node is attempting to transmit
to the destination

Moral of the story: The link reversal algorithm in
[Gafni81] does not dictate a proactive or reactive
response to link failure/repair

Decision on reactive/proactive behavior should be
made based on environment under consideration
143
So far ...

All nodes had identical responsibilities

Some schemes propose giving special
responsibilities to a subset of nodes
“Core” based schemes assign additional tasks to nodes
belonging to the “core
Clustering schemes assign additional tasks to cluster
“leaders”

Not discussed further in this tutorial
144
Proactive Protocols
145
Proactive Protocols

Most of the schemes discussed so far are reactive

Proactive schemes based on distance-vector and
link-state mechanisms have also been proposed
146
Link State Routing [Huitema95]

Each node periodically floods status of its links

Each node re-broadcasts link state information
received from its neighbor

Each node keeps track of link state information
received from other nodes

Each node uses above information to determine next
hop to each destination
147
Optimized Link State Routing (OLSR)
[Jacquet00ietf,Jacquet99Inria]

The overhead of flooding link state information is
reduced by requiring fewer nodes to forward the
information

A broadcast from node X is only forwarded by its
multipoint relays

Multipoint relays of node X are its neighbors such
that each two-hop neighbor of X is a one-hop
neighbor of at least one multipoint relay of X
Each node transmits its neighbor list in periodic beacons, so
that all nodes can know their 2-hop neighbors, in order to
choose the multipoint relays
148
Optimized Link State Routing (OLSR)

Nodes C and E are multipoint relays of node A
F
B
A
C
G
J
E
H
K
D
Node that has broadcast state information from A
149
Optimized Link State Routing (OLSR)

Nodes C and E forward information received from A
F
B
A
C
G
J
E
H
K
D
Node that has broadcast state information from A
150
Optimized Link State Routing (OLSR)


Nodes E and K are multipoint relays for node H
Node K forwards information received from H
E has already forwarded the same information once
F
B
A
C
G
J
E
H
K
D
Node that has broadcast state information from A
151
OLSR

OLSR floods information through the multipoint relays

The flooded itself is fir links connecting nodes to
respective multipoint relays

Routes used by OLSR only include multipoint relays
as intermediate nodes
152
Destination-Sequenced Distance-Vector
(DSDV) [Perkins94Sigcomm]

Each node maintains a routing table which stores
next hop towards each destination
a cost metric for the path to each destination
a destination sequence number that is created by the
destination itself
Sequence numbers used to avoid formation of loops

Each node periodically forwards the routing table to
its neighbors
Each node increments and appends its sequence number
when sending its local routing table
This sequence number will be attached to route entries
created for this node
153
Destination-Sequenced Distance-Vector
(DSDV)

Assume that node X receives routing information
from Y about a route to node Z
X

Y
Z
Let S(X) and S(Y) denote the destination sequence
number for node Z as stored at node X, and as sent
by node Y with its routing table to node X,
respectively
154
Destination-Sequenced Distance-Vector
(DSDV)

Node X takes the following steps:
X
Y
Z
If S(X) > S(Y), then X ignores the routing information
received from Y
If S(X) = S(Y), and cost of going through Y is smaller than
the route known to X, then X sets Y as the next hop to Z
If S(X) < S(Y), then X sets Y as the next hop to Z, and S(X)
is updated to equal S(Y)
155
Hybrid Protocols
156
Zone Routing Protocol (ZRP) [Haas98]
Zone routing protocol combines

Proactive protocol: which pro-actively updates
network state and maintains route regardless of
whether any data traffic exists or not

Reactive protocol: which only determines route to a
destination if there is some data to be sent to the
destination
157
ZRP

All nodes within hop distance at most d from a node
X are said to be in the routing zone of node X

All nodes at hop distance exactly d are said to be
peripheral nodes of node X’s routing zone
158
ZRP

Intra-zone routing: Pro-actively maintain state
information for links within a short distance from any
given node
Routes to nodes within short distance are thus maintained
proactively (using, say, link state or distance vector protocol)

Inter-zone routing: Use a route discovery protocol for
determining routes to far away nodes. Route
discovery is similar to DSR with the exception that
route requests are propagated via peripheral nodes.
159
ZRP: Example with
Zone Radius = d = 2
S performs route
discovery for D
B
S
A
F
Denotes route request
C
E
D
160
ZRP: Example with d = 2
S performs route
discovery for D
B
S
A
F
Denotes route reply
C
E
D
E knows route from E to D,
so route request need not be
161
forwarded to D from E
ZRP: Example with d = 2
S performs route
discovery for D
B
S
A
C
F
Denotes route taken by Data
E
D
162
Landmark Routing (LANMAR) for MANET with
Group Mobility [Pei00Mobihoc]

A landmark node is elected for a group of nodes that
are likely to move together

A scope is defined such that each node would
typically be within the scope of its landmark node

Each node propagates link state information
corresponding only to nodes within it scope and
distance-vector information for all landmark nodes
Combination of link-state and distance-vector
Distance-vector used for landmark nodes outside the scope
No state information for non-landmark nodes outside scope
maintained
163
LANMAR Routing to Nodes Within Scope

Assume that node C is within scope of node A
H
C
A

B
G
D
E
F
Routing from A to C: Node A can determine next hop
to node C using the available link state information
164
LANMAR Routing to Nodes Outside Scope


Routing from node A to F which is outside A’s scope
Let H be the landmark node for node F
H
C
A


B
G
D
E
F
Node A somehow knows that H is the landmark for C
Node A can determine next hop to node H using the
available distance vector information
165
LANMAR Routing to Nodes Outside Scope

Node D is within scope of node F
H
C
A


B
G
D
E
F
Node D can determine next hop to node F using link
state information
The packet for F may never reach the landmark node
H, even though initially node A sends it towards H
166

LANMAR scheme uses node identifiers as landmarks

Anchored Geodesic Scheme [LeBoudec00] uses
geographical regions as landmarks
167
Routing

Protocols discussed so far find/maintain a route
provided it exists

Some protocols attempt to ensure that a route exists
by
Power Control [Ramanathan00Infocom]
Limiting movement of hosts or forcing them to take detours
[Reuben98thesis]
168
Power Control

Protocols discussed so far find a route, on a given network
topology

Some researchers propose controlling network topology by
transmission power control to yield network properties which
may be desirable [Ramanathan00Infocom]
Such approaches can significantly impact performance at several
layers of protocol stack

[Wattwnhofer00Infocom] provides a distributed mechanism for
power control which allows for local decisions, but guarantees
global connectivity
Each node uses a power level that ensures that the node has at
least one neighbor in each cone with angle 2p/3
169
Some Variations
170
Power-Aware Routing
[Singh98Mobicom,Chang00Infocom]
Define optimization criteria as a function of energy
consumption. Examples:

Minimize energy consumed per packet

Minimize time to network partition due to energy
depletion

Maximize duration before a node fails due to energy
depletion
171
Power-Aware Routing [Singh98Mobicom]

Assign a weight to each link

Weight of a link may be a function of energy
consumed when transmitting a packet on that link, as
well as the residual energy level
low residual energy level may correspond to a high cost

Prefer a route with the smallest aggregate weight
172
Power-Aware Routing
Possible modification to DSR to make it power aware
(for simplicity, assume no route caching):

Route Requests aggregate the weights of all
traversed links

Destination responds with a Route Reply to a Route
Request if
it is the first RREQ with a given (“current”) sequence
number, or
its weight is smaller than all other RREQs received with the
current sequence number
173
Preemptive Routing [Goff01MobiCom]

Add some proactivity to reactive routing protocols
such as DSR and AODV

Route discovery initiated when it appears that an
active route will break in the near future

Initiating route discover before existing route breaks
reduces discovery latency
174
Performance of Unicast Routing in MANET

Several performance comparisons
[Broch98Mobicom,Johansson99Mobicom,Das00Infocom,
Das98ic3n]

We will discuss performance issue later in the tutorial
175
Address Auto-Configuration
176
Address Auto-configuration

Auto-configuration important for autonomous
operation of an ad hoc network

IPv4 and IPv6 auto-configuration mechanisms have
been proposed
• Need to be adapted for ad hoc networks
177
Auto-Configuration in
Ad Hoc Networks

Worst case network delays may be unknown, or
highly variable

Partitions may occur, and merge
178
Duplicate Address Detection
in Ad Hoc Networks


Several proposals
One example [Perkins]:
Host picks an address randomly
Host performs route discovery for the chosen address
If a route reply is received, address duplication is detected
179
Example:
Initially Partitioned Network
D’s packets for address a routed to A
180
Merged Network

Duplicate address detection (DAD) important To avoid
misrouting
181
Strong DAD

Detect duplicate addresses within t seconds

Not possible to guarantee strong DAD in presence of
unbounded delays
May occur due to partitions
Even when delays are bounded, bound may be difficult to
calculate
Unknown network size
•
182
DAD

Strong DAD impossible with unbounded delay

How to achieve DAD ?
183
Design Principle

If you cannot solve a problem
Change the problem
184
Weak DAD [Vaidya02MobiHoc]
Packets from a given host to a given address
should be routed to the same destination,
despite duplication of the address
185
Example:
Initially Partitioned Network
D’s packets for address a routed to A
186
Merged Network:
Acceptable Behavior
with Weak DAD
Packets from D
to address a
still routed to
host A
187
Merged Network:
Unacceptable behavior
Packets from D
to address a
routed to
host K instead
of A
188
Weak DAD: Implementation

Integrate duplicate address detection with route
maintenance
189
Weak DAD with Link State Routing

Each host has a unique (with high probability) key
May include MAC address, serial number, …
May be large in size

In all routing-related packets (link state updates) IP
addresses tagged by keys
 (IP, key) pair
190
Weak DAD with Link State Routing

Address duplication not always detected

Duplication detected before misrouting can occur

Weak
 Reliable, but potentially delayed, DAD
191
Link State Routing (LSR): Example
192
Weak DAD with LSR
193
Weak DAD with LSR
X
Host X with key K_x joins
and choose IP_A
(address duplication)
194
Weak DAD with LSR
If host D receives a link state
update containing (IP_A, K_x),
host D detects duplication of
address IP_A
Two pairs with identical IP
address but distinct keys imply
duplication
195
Just-in-Time DAD

Duplication detected before routing tables could be
mis-configured
196
Higher Layer Interaction

Higher layers interaction may result in undesirable
behavior
197
Example
Q discovers service Foo at address a
198
Example: Networks merge
Node A
performs
service discovery
for Foo, and
learns from Q
that Foo is
available at
address a
199
Example: Networks merge
Node A’s
packets to a
are delivered to M
R provides service
Foo not M
200
Enhanced Weak DAD

If the status of host A above the network layer
depends on state of host B
(State A  state B)
 then network layer of host A should be aware of (IP, key)
pairs known to B
201
Enhanced Weak DAD

Works despite upper layer interaction
202
Weak DAD: Other Issues

Duplicate MAC addresses within two hops of each
other bad
• Need a duplicate MAC address detection scheme

Network layers performing unicasts using
multicast/flooding

Limited-time address leases

DAD with other routing protocols
Possible. Paper also discusses DSR.
203
Summary

Strong DAD – Not always possible

Weak DAD feasible
Combines DAD with route maintenance

Overhead of weak DAD
Expected to be low, but unknown presently
204
Capacity of Ad Hoc Networks
205
Capacity of Fixed Ad Hoc Networks
[Gupta00it]

n nodes in area A transmitting at W bits/sec using a
fixed range (distance between a random pair of
nodes is O(sqrt(n))

Bit-distance product that can be transported by the
network per second is
Q ( W sqrt (A n) )

Throughput per node
Q(W
/ sqrt (n) )
206
Capacity of Mobile Ad Hoc Networks
[Grossglauser01Infocom]


Assume random motion
Any two nodes become neighbors once in a while

Each node assumed sender for one session, and
destination for another session

Relay packets through at most one other node
Packet go from S to D directly, when S and D are neighbors,
or from S to a relay and the the relay to D, when each pair
becomes neighbor respectively

Throughput of each session is O(1)
Independent of n
207
Continues from last slide …

Delay in packet delivery can be large if O(1)
throughput is to be achieved
Delay incurred waiting for the destination to arrive close to a
relay or the sender

Trade-off between delay and throughput
208
Measured Capacity [Li01MobiCom]

Confirms intuition

In fixed networks, capacity is higher if average
distance between source-destination pairs is small
209
Measured Scaling Law
[Gupta0]

Measured in static networks

Throughput declines worse with n than theoretically
predicted

Existing MAC protocols unable to exploit “parallelism”
in channel access
210
Capacity

How to design MAC and routing protocols to
approach theoretical capacity ?

Open problem
211
Medium Access Control Protocols
212
Medium Access Control

Wireless channel is a shared medium

Need access control mechanism to avoid
interference

MAC protocol design has been an active area of
research for many years [Chandra00survey]
213
MAC: A Simple Classification
Wireless
MAC
Centralized
Distributed
Guaranteed
or
controlled
access
Random
access
This
tutorial
214
This tutorial

Mostly focus on random access protocols

Not a comprehensive overview of MAC protocols

Provides discussion of some example protocols
215
Hidden Terminal Problem [Tobagi75]




Node B can communicate with A and C both
A and C cannot hear each other
When A transmits to B, C cannot detect the
transmission using the carrier sense mechanism
If C transmits, collision will occur at node B
A
B
C
216
Busy Tone [Tobagi75,Haas98]

A receiver transmits busy tone when receiving data

All nodes hearing busy tone keep silent

Avoids interference from hidden terminals

Requires a separate channel for busy tone
217
MACA Solution for Hidden Terminal Problem
[Karn90]

When node A wants to send a packet to node B,
node A first sends a Request-to-Send (RTS) to A

On receiving RTS, node A responds by sending
Clear-to-Send (CTS), provided node A is able to
receive the packet

When a node (such as C) overhears a CTS, it keeps
quiet for the duration of the transfer
Transfer duration is included in RTS and CTS both
A
B
C
218
Reliability

Wireless links are prone to errors. High packet loss
rate detrimental to transport-layer performance.

Mechanisms needed to reduce packet loss rate
experienced by upper layers
219
A Simple Solution to Improve Reliability

When node B receives a data packet from node A,
node B sends an Acknowledgement (Ack). This
approach adopted in many protocols
[Bharghavan94,IEEE 802.11]

If node A fails to receive an Ack, it will retransmit the
packet
A
B
C
220
IEEE 802.11 Wireless MAC

Distributed and centralized MAC components
Distributed Coordination Function (DCF)
Point Coordination Function (PCF)

DCF suitable for multi-hop ad hoc networking

DCF is a Carrier Sense Multiple Access/Collision
Avoidance (CSMA/CA) protocol
221
IEEE 802.11 DCF

Uses RTS-CTS exchange to avoid hidden terminal
problem
Any node overhearing a CTS cannot transmit for the
duration of the transfer

Uses ACK to achieve reliability

Any node receiving the RTS cannot transmit for the
duration of the transfer
To prevent collision with ACK when it arrives at the sender
When B is sending data to C, node A will keep quite
A
B
C
222
Collision Avoidance

With half-duplex radios, collision detection is not possible

CSMA/CA: Wireless MAC protocols often use collision
avoidance techniques, in conjunction with a (physical or virtual)
carrier sense mechanism

Carrier sense: When a node wishes to transmit a packet, it first
waits until the channel is idle.

Collision avoidance: Nodes hearing RTS or CTS stay silent for
the duration of the corresponding transmission. Once channel
becomes idle, the node waits for a randomly chosen duration
before attempting to transmit.
223
IEEE 802.11
RTS = Request-to-Send
RTS
A
B
C
D
E
F
Pretending a circular range
224
IEEE 802.11
RTS = Request-to-Send
RTS
A
B
C
D
E
F
NAV = 10
225
NAV = remaining duration to keep quiet
IEEE 802.11
CTS = Clear-to-Send
CTS
A
B
C
D
E
F
226
IEEE 802.11
CTS = Clear-to-Send
CTS
A
B
C
D
E
F
NAV = 8
227
IEEE 802.11
•DATA packet follows CTS. Successful data reception
acknowledged using ACK.
DATA
A
B
C
D
E
F
228
IEEE 802.11
ACK
A
B
C
D
E
F
229
IEEE 802.11
Reserved area
ACK
A
B
C
D
E
F
230
IEEE 802.11
Carrier sense
range
Interference
“range”
DATA
A
B
C
D
E
F
Transmit “range”
231
CSMA/CA





Physical carrier sense, and
Virtual carrier sense using Network Allocation Vector
(NAV)
NAV is updated based on overheard
RTS/CTS/DATA/ACK packets, each of which
specified duration of a pending transmission
Nodes stay silent when carrier sensed
(physical/virtual)
Backoff intervals used to reduce collision probability
232
Backoff Interval

When transmitting a packet, choose a backoff interval
in the range [0,cw]
cw is contention window

Count down the backoff interval when medium is idle
Count-down is suspended if medium becomes busy

When backoff interval reaches 0, transmit RTS
233
DCF Example
B1 = 25
B1 = 5
wait
data
data
B2 = 20
cw = 31
wait
B2 = 15
B2 = 10
B1 and B2 are backoff intervals
at nodes 1 and 2
234
Backoff Interval

The time spent counting down backoff intervals is a
part of MAC overhead

Choosing a large cw leads to large backoff intervals
and can result in larger overhead

Choosing a small cw leads to a larger number of
collisions (when two nodes count down to 0
simultaneously)
235

Since the number of nodes attempting to transmit
simultaneously may change with time, some
mechanism to manage contention is needed

IEEE 802.11 DCF: contention window cw is chosen
dynamically depending on collision occurrence
236
Binary Exponential Backoff in DCF

When a node fails to receive CTS in response to its
RTS, it increases the contention window
cw is doubled (up to an upper bound)

When a node successfully completes a data transfer,
it restores cw to Cwmin

cw follows a sawtooth curve
237
MILD Algorithm in MACAW [Bharghavan94]

When a node successfully completes a transfer,
reduces cw by 1
In 802.11 cw is restored to cwmin
In 802.11, cw reduces much faster than it increases
MACAW: cw reduces slower than it increases
Exponential Increase Linear Decrease

MACAW can avoid wild oscillations of cw when large
number of nodes contend for the channel
238
Alternative Contention Resolution Mechanism
[Hiperlan]

Elimination phase
A node transmits a burst for a random number
(geometrically distributed) of slots
If medium idle at the end of the burst, go to yield phase, else
give up until next round

Yield phase
Stay silent for a random number (geometrical distributed) of
slots
If medium still silent, transmit
239
Contention Resolution Overhead

Channel contention resolved using backoff
Nodes choose random backoff interval from [0, CW]
Count down for this interval before transmission

Backoff and (optional) RTS/CTS handshake before
transmission of data packet
Random
backoff
RTS/CTS
Data Transmission/ACK
240
Inefficiency of IEEE 802.11

Backoff interval should be chosen appropriately for
efficiency

Backoff interval with 802.11 far from optimum
241
Observation

Backoff and RTS/CTS handshake are unproductive:
Do not contribute to throughput
Unproductive
Random
backoff
RTS/CTS
Data Transmission/ACK
242
Observation

Terry Todd observed that if a protocol has a
“bandwidth-independent” overhead it is possible to
improve performance by moving the bandwidthindependent overhead to a narrowband channel

Pipelining motivated by these observations
243
Pipelining [Yan02techrep]

Two stage pipeline:
Random backoff and RTS/CTS handshake
Data transmission and ACK

“Total” pipelining: Resolve contention completely in
stage 1
Random
backoff
Stage1
RTS/CTS
Data Transmission/ACK
Stage2
244
How to Pipeline ?

Use two channels
Control Channel: Random backoff and RTS/CTS handshake
Data Channel: Data transmission and ACK
Random
backoff
RTS/CTS
Random
backoff
RTS/CTS
Data Transmission/ACK
Random
backoff
RTS/CTS
Data Transmission/ACK
245
Pipelining

Pipelining works well only if two stages are balanced!
Random
backoff
RTS/CTS
Random
backoff
RTS/CTS
Random
backoff
RTS/CTS
Control Channel
Data Transmission/ACK
Data Transmission/ACK
Data Channel
246
Pipelining

Length of stage 1 depends on:
Control channel bandwidth
The random backoff duration
The number of collisions occurred

Length of stage 2 depends on:
Data channel bandwidth
The data packet size
247
How much bandwidth does control channel
require?

If small, then
RTS/CTS takes very long time.
Collision detection is slow

If large, then
The portion of channel bandwidth used for productive data
packet transmission is reduced
Total bandwidth is fixed!
248
Difficulty with Total Pipelining

The optimum division of channel bandwidth varies
with contention level and data packet size

Performance with inappropriate bandwidth division
could be even worse than 802.11 DCF
249
How to get around the issue of bandwidth division ?
250
Partial Pipelining

Only partially resolve channel contention in stage 1

Since no need to completely resolve contention, the
length of stage 1 can be elastic to match the length of
stage 2
251
Modified Two Stage Pipeline


Stage 1: Random backoff phase 1
Stage 2: Random backoff phase 2, RTS/CTS
handshake and Data/ACK transmission
Backoff phase 1
Stage1
Backoff
phase 2
RTS/CTS
Data/ACK
Stage2
252

Still use two channels
Narrow Band Busy Tone Channel:
• Random backoff phase 1
Data Channel: Random backoff phase 2, RTS/CTS
handshake and Data/ACK
Random backoff phase 1
Random backoff phase 1
Random backoff phase 1
Backoff
RTS/CTS Data/ACK
phase 2
Backoff
RTS/CTS Data/ACK
phase 2
253
Random Backoff Phase 1

Each Station maintains a counter for random backoff
phase 1

The stations, which count to zero first, send a busy
tone to claim win in stage 1
Multiple winners are possible

Other stations know they lost on sensing a busy tone
254
Gain over total pipelining?

No packets transmitted on busy tone channel
bandwidth can be small
the difficulty of deciding optimum bandwidth division in “total
pipelining” is avoided

Length of stage 1 is elastic so the two stages can be
kept balanced
255
Benefits of Partial Pipeline

Only winners of stage 1 can contend channel in stage 2
reduces the data channel contention
reduces collision probability on the data channel
Stage 1
Stage 2
256
Sounds like HIPERLAN/1?
HIPERLAN / 1 (no pipelining)
Elimination
Stage
Yield Stage
Data Transmission
Partial Pipelining
Random backoff phase 1
Random backoff phase 1
Random backoff phase 1
Backoff
RTS/CTS Data/ACK
phase 2
Backoff
RTS/CTS Data/ACK
phase 2
257
Benefits of Partial Pipeline

Because of pipelining, stages 1 and 2 proceed
in parallel. Stage 1 costs little except for a narrow
band busy tone channel
Partial Pipelining
Random backoff phase 1
Random backoff phase 1
Random backoff phase 1
Backoff
RTS/CTS Data/ACK
phase 2
Backoff
RTS/CTS Data/ACK
phase 2
258
Benefits of Partial Pipeline

By migrating most of the backoff to busy tone
channel,bandwidth cost of random backoff is reduced
Cost of backoff = Channel bandwidth * backoff duration
Area = cost
of backoff
Data Channel
Bandwidth
Busy Tone
Channel Bandwidth
Using IEEE 802.11
DSSS, the backoff
duration could be
several milliseconds
Backoff Duration
259
Results of Partial Pipelining

Improved throughput and stability over 802.11 DCF
Partial
Pipelining
802.11 DCF
260
Fairness
261
Fairness Issue

Many definitions of fairness plausible

Simplest definition: All nodes should receive equal
bandwidth
A
B
Two flows
C
D
262
Fairness Issue

Assume that initially, A and B both choose a backoff
interval in range [0,31] but their RTSs collide

Nodes A and B then choose from range [0,63]
Node A chooses 4 slots and B choose 60 slots
After A transmits a packet, it next chooses from range [0,31]
It is possible that A may transmit several packets before B
transmits its first packet
A
B
Two flows
C
D
263
Fairness Issue

Unfairness occurs when one node has backed off
much more than some other node
A
B
Two flows
C
D
264
MACAW Solution for Fairness

When a node transmits a packet, it appends the cw
value to the packet, all nodes hearing that cw value
use it for their future transmission attempts

Since cw is an indication of the level of congestion in
the vicinity of a specific receiver node, MACAW
proposes maintaining cw independently for each
receiver

Using per-receiver cw is particularly useful in multihop environments, since congestion level at different
receivers can be very different
265
Another MACAW Proposal

For the scenario below, when node A sends an RTS
to B, while node C is receiving from D, node B
cannot reply with a CTS, since B knows that D is
sending to C

When the transfer from C to D is complete, node B
can send a Request-to-send-RTS to node A
[Bharghavan94Sigcomm]
Node A may then immediately send RTS to node B
A
B
C
D
266

This approach, however, does not work in the
scenario below
Node B may not receive the RTS from A at all, due to
interference with transmission from C
A
B
C
D
267
Weighted Fair Queueing [Keshav97book]

Assign a weight to each node

Bandwidth used by each node should be proportional
to the weight assigned to the node
268
Distributed Fair Scheduling (DFS)
[Vaidya00Mobicom]

A fully distributed algorithm for achieving weighted
fair queueing

Chooses backoff intervals proportional to
(packet size / weight)

DFS attempts to mimic the centralized Self-Clocked
Fair Queueing algorithm [Golestani]

Works well on a LAN
269
Distributed Fair Scheduling (DFS)
B1 = 10
B1 = 15
wait
B1 = 5
wait
Collision !
data
B2 = 5
data
B2 = 5
B2 = 5
Weight of node 1 = 1
Weight of node 2 = 3
B1 = 15 (DFS actually picks a random value
with mean 15)
Assume equal
packet size
B2 = 5
(DFS picks a value with mean 5)
270
Impact of Collisions

After collision resolution, either node 1 or node 2 may
transmit a packet

The two alternatives may have different fairness
properties (since collision resolution can result in
priority inversion)
271
Distributed Fair Scheduling (DFS)
B1 = 10
B1 = 10
wait
wait
data
B2 = 5
B1 = 5
wait
data
B2 = 5
da
data
B2 = 5
Collision resolution
272
Distributed Fair Scheduling

DFS uses randomization to reduce collisions
Alleviates negative impact of synchronization

DFS also uses a shifted contention window for
choosing initial backoff interval
Reduces priority inversion (which leads to short-term
unfairness)
802.11
0
31
0
31
DFS
273
DFS

Due to large cw, DFS can potentially yield lower
throughput than IEEE 802.11
trade-off between fairness and throughput

On multi-hop network, properties of DFS still need to
be characterized

Fairness in multi-hop case affected by hidden
terminals
May need use of a copying technique, analogous to window
copying in MACAW, to share some protocol state
274
Fairness in Multi-Hop Networks

Several definitions of fairness
[Ozugur98,Vaidya99MSR,Luo00Mobicom,
Nandagopal00Mobicom]

Hidden terminals make it difficult to achieve a desired
notion of fairness
275
Estimation-Based Fair MAC
[Bansou00MobiHoc]

Attempts to equalize throughput/weight ratio for all
nodes

Two parts of the algorithm
Fair share estimation
Window adjustment

Each node estimates how much bandwidth (W) it is
able to use, and the amount of bandwidth used by
each station in its vicinity
Estimation based on overheard RTS, CTS, DATA packets
276
Estimation-Based Fair MAC

Fair share estimation: Node estimates how much
bandwidth (Wi) it is able to use, and the amount of
bandwidth (Wo) used by by all other neighbors
combined
Estimation based on overheard RTS, CTS, DATA packets
277
Estimation-Based Fair MAC

Define:
Ti = Wi / weight of i
To = Wo / weight assigned to the group of neighbors of i
Fairness index = Ti / To

Window adjustment:
If fairness index is too large, cw = cw * 2
Else if fairness index is too small, cw = cw / 2
Else no change to cw (contention window)
278
Proportional Fair Contention Resolution (PFCR)
[Nandagopal00Mobicom]

Proportional fairness: Allocate bandwidth Ri to node i
such that any other allocation Si has the following
property
Si

(Si-Ri) / Ri < 0
Link access probability is dynamically changed
depending on success/failure at transmitting a packet
On success: Link access probability is increased by an
additive factor a
On failure: Link access probability is decreased by a
multiplicative factor (1-b)
279
Sender-Initiated Protocols

The protocols discussed so far are sender-initiated
protocols

The sender initiates a packet transfer to a receiver
280
Receive-Initiated Mechanism
[Talucci97,Garcia99]

In most protocols, sender initiates a transfer

Alternatively, a receiver may send a
Ready-To-Receive (RTR) message to a sender
requesting it to being a packet transfer

Sender node on receiving the RTR transmits data

How does a receiver determine when to poll a sender
with RTR?
Based on history, and prediction of traffic from the sender
281
Energy Conservation
282
Energy Conservation

Since many mobile hosts are operated by batteries,
MAC protocols which conserve energy are of interest

Two approaches to reduce energy consumption
Power save: Turn off wireless interface when desirable
Power control: Reduce transmit power
283
Power Aware Multi-Access Protocol (PAMAS)
[Singh98]

A node powers off its radio while a neighbor is
transmitting to someone else
Node A sending to B
Node C stays powered off
B
A
C
284
Power Aware Multi-Access Protocol (PAMAS)

What should node C do when it wakes up and finds
that D is transmitting to someone else
C does not know how long the transfer will last
Node D sending to E
Node A sending to B
C stays powered off
B
A
C
D
C wakes up and
finds medium busy
E
285
PAMAS

PAMAS uses a control channel separate from the
data channel

Node C on waking up performs a binary probe to
determine the length of the longest remaining transfer
C sends a probe packet with parameter L
All nodes which will finish transfer in interval [L/2,L] respond
Depending on whether node C see silence, collision, or a
unique response it takes varying actions

Node C (using procedure above) determines the
duration of time to go back to sleep
286
Disadvantages of PAMAS

Use of a separate control channel

Nodes have to be able to receive on the control
channel while they are transmitting on the data
channel
And also transmit on data and control channels
simultaneously

A node (such as C) should be able to determine
when probe responses from multiple senders collide
287
Another Proposal in PAMAS

To avoid the probing, a node should switch off the
interface for data channel, but not for the control
channel (which carries RTS/CTS packets)

Advantage: Each sleeping node always know how
long to sleep by watching the control channel

Disadvantage: This may not be useful when
hardware is shared for the control and data channels
It may not be possible turn off much hardware due to the
sharing
288
Power Save in IEEE 802.11 Ad Hoc Mode

Time is divided into beacon intervals
ATIM
window
Beacon interval

Each beacon interval begins with an ATIM window
ATIM =
289
Power Save in IEEE 802.11 Ad Hoc Mode

If host A has a packet to transmit to B, A must send
an ATIM Request to B during an ATIM Window

On receipt of ATIM Request from A, B will reply by
sending an ATIM Ack, and stay up during the rest of
the beacon interval

If a host does not receive an ATIM Request during an
ATIM window, and has no pending packets to
transmit, it may sleep during rest of the beacon
interval
290
Power Save in IEEE 802.11 Ad Hoc Mode
Node A
ATIM
Req
ATIM Data
Ack
Ack
Node B
Node C
Sleep
291
Power Save in IEEE 802.11 Ad Hoc Mode

Size of ATIM window and beacon interval affects
performance [Woesner98]

If ATIM window is too large, reduction in energy
consumption reduced
Energy consumed during ATIM window

If ATIM window is too small, not enough time to send
ATIM request
292
Power Save in IEEE 802.11 Ad Hoc Mode

How to choose ATIM window dynamically?
Based on observed load [Jung02infocom]

How to synchronize hosts?
If two hosts’ ATIM windows do not overlap in time, they
cannot exchange ATIM requests
Coordination requires that each host stay awake long
enough (at least periodically) to discover out-of-sync
neighbors [Tseng02infocom]
ATIM
ATIM
293
Impact on Upper Layers

If each node uses the 802.11 power-save
mechanism, each hop will require one beacon
interval
This delay could be intolerable

Allow upper layers to dictate whether a node should
enter the power save mode or not [Chen01mobicom]
294
Power Save Using Wake-Up Channels
295
Motivation

Sleep mode power consumption << Idle power consumption
Power Characteristics for a Mica2 Mote Sensor
296
Design Alternatives

Synchronous: Once a host enters sleep mode, it
wakes up at a pre-determined time
• Timer-based

Asynchronous: A sleeping host can be woken up at
any time by a neighbor

Hybrid: Synchronous + Asynchronous
297
Using Wake-up Radio [Miller04WCNC]

Add second, low-power radio to wakeup neighbors
on-demand

Low-power wake-up can be achieved using
Simpler hardware with a lower bit-rate and/or less decoding
capability, or
A periodic duty cycle (e.g., as in STEM [UCLA]) using a
“normal” radio
– Latter approach used in the illustration here
298
Actions of a Sleeping Host

Periodically listen to a wake-up channel

If wake-up channel sensed busy:
Turn on data radio
Receive a “filter” packet on data radio
If filter intended for another host, go back to sleep
• Duty cycle affects energy consumption
299
Actions of a Sender Host

Transmit a wake-up signal “long enough” if the
intended receiver is expected to be sleeping

Transmit a filter packet specifying intended receiver

Transmit data to the receiver
300
Purely Asynchronous Mechanism

In a purely asynchronous approach, each packet
burse is preceded by a “wake-up” signal

Might wake-up too many hosts near the transmitter –
referred as “full” wakeup
 Energy cost
301
Add a Synchronous Component

Each sleeping host will wake-up after a pre-defined
interval of time (“timeout”)
Referred as “triggered” wakeup

If a transmitter cannot wait until then, it may send a
wake-up signal
Send wake-up signal if queue size exceeds threshold L or a
delay bound

Timeout is computed based on recent traffic rate
302
Timeout for Triggered Wakeups

If too small, host may wake-up when there are no
packets pending for it

If too large, too many “full” wakeups
303
Power Save Protocol [Miller04WCNC]
304
Energy Conservation

Power save

Power control
305
Power Control
Power control has two potential benefit

Reduced interference & increased spatial reuse

Energy saving
306
Power Control

When C transmits to D at a high power level, B
cannot receive A’s transmission due to interference
from C
A
B
C
D
307
Power Control

If C reduces transmit power, it can still communicate
with D
• Reduces energy consumption at node C
• Allows B to receive A’s transmission (spatial reuse)
A
B
C
D
308
Power Control

a
Received power level is proportional to 1/d , a >= 2

If power control is utilized, energy required to transmit
to a host at distance d is proportional to
a
d + constant

Shorter hops typically preferred for energy consumption
(depending on the constant) [Rodoplu99]
Transmit to C from A via B, instead of directly from A to C
A
B
C
309
Power Control with 802.11

A
Transmit RTS/CTS/DATA/ACK at least power level
needed to communicate with the received
B
C
D

A/B do not receive RTS/CTS from C/D. Also do not
sense D’s data transmission

B’s transmission to A at high power interferes with
reception of ACK at C
310
A Plausible Solution

RTS/CTS at highest power, and DATA/ACK at smallest
necessary power level
Data sensed
A
B
C
D
Data
Interference range



RTS
Ack
A cannot sense C’s data transmission, and may transmit DATA
to some other host
This DATA will interfere at C
This situation unlikely if DATA transmitted at highest power level
Interference range ~ sensing range
311

Transmitting RTS at the highest power level also
reduces spatial reuse

Nodes receiving RTS/CTS have to defer
transmissions
312
Modification to Avoid Interference


Transmit RTS/CTS at highest power level, DATA/ACK
at least required power level
Increase DATA power periodically so distant hosts
can sense transmission [Jung02tech]
Power
level

Need to be able to change power level rapidly
313
Caveat

Energy saving by power control is limited to savings
in transmit energy

Other energy costs may not change, and may
represent a significant fraction of total energy
consumption
314
Power Controlled Multiple Access (PCMA)
[Monks01infocom]

If receiver node R can tolerate noise E, it sends a
busy tone at power level C/E, where C is an
appropriate constant

When some node X receives a busy-tone a power
level Pr, it may transmit at power level Pt <= C/Pr
busy tone
data
S
R
X
Pt
C/E
Y
315
Power Controlled Multiple Access (PCMA)
[Monks01infocom]

If receiver node R can tolerate noise E, it sends a
busy tone at power level C/E, where C is an
appropriate constant

When some node X receives a busy-tone a power
level Pr, it may transmit at power level Pt <= C/Pr

Explanation:
Gain of channel RX = gain of channel XR = g
Busy tone signal level at X = Pr = g * C / E
Node X may transmit at level = Pt = C/Pr = E/g
Interference received by R = Pt * g = E
316
PCMA

Advantage
Allows higher spatial reuse, as well as power saving using
power control

Disadvantages:
Need a separate channel for the busy tone
Since multiple nodes may transmit the busy tones
simultaneously, spatial reuse is less than optimal
317
Small Addresses Save Energy
[Schurgers01mobihoc]

In sensor networks, packet sizes are small, and MAC addresses
may be a substantial fraction of the packet

Observation: MAC addresses need only be unique within two
hops
E1
G0
D3
A2
C0
F2
B1

Fewer addresses are sufficient: Address size can be smaller.
[Schurgers00mobihoc] uses Huffman coding to assign variable
size encoding to the addresses
Energy consumption reduced due to smaller addresses

318
Adaptive Modulation
319
Adaptive Modulation

Channel conditions are time-varying

Received signal-to-noise ratio changes with time
A
B
320
Adaptive Modulation


Multi-rate radios are capable of transmitting at
several rates, using different modulation schemes
Choose modulation scheme as a function of channel
conditions
Modulation schemes provide
a trade-off between
throughput and range
Throughput
Distance
321
Adaptive Modulation

If physical layer chooses the modulation scheme
transparent to MAC
MAC cannot know the time duration required for the transfer

Must involve MAC protocol in deciding the
modulation scheme
Some implementations use a sender-based scheme for this
purpose [Kamerman97]
Receiver-based schemes can perform better
322
Sender-Based “Autorate Fallback”
[Kamerman97]

Probing mechanisms

Sender decreases bit rate after X consecutive
transmission attempts fail

Sender increases bit rate after Y consecutive
transmission attempt succeed
323
Autorate Fallback

Advantage
Can be implemented at the sender, without making any
changes to the 802.11 standard specification

Disadvantage
Probing mechanism does not accurately detect channel
state
Channel state detected more accurately at the receiver
Performance can suffer
Since the sender will periodically try to send at a rate
higher than optimal
Also, when channel conditions improve, the rate is not
increased immediately
•
•
324
Receiver-Based Autorate MAC
[Holland01mobicom]

Sender sends RTS containing its best rate estimate

Receiver chooses best rate for the conditions and
sends it in the CTS

Sender transmits DATA packet at new rate

Information in data packet header implicitly updates
nodes that heard old rate
325
Receiver-Based Autorate MAC Protocol
C
A
RTS (2 Mbps)
B
CTS (1 Mbps)
Data (1 Mbps)
D
NAV updated
using rate
specified in the
data packet
326
Multiple Channels
327
Multiple Channels

Multiple channels in ad hoc networks: typically
defined FDMA

TDMA requires time synchronization among hosts in
ad hoc network
Difficult

Many MAC protocols have been proposed
328
Multi-Channel MAC: A simple approach

Divide bandwidth into multiple channels

Choose any one of the idle channels

Use a single-channel protocol on the chosen channel
for instance, 802.11
329
Multi-Channel MAC with Soft Reservation
[Nasipuri00]

Similar to the simple scheme, channel used recently
for a successful transmission preferred

Tends to “reserve” channels
330
Another Protocol

Use one (control) channel for RTS/CTS and
remaining (data) channels for DATA/ACK

Each host maintains NAV table, with one entry for
each data channel

Sender sends RTS to destination, specifying the
channels that are free per sender’s table

Receiver replies with CTS specifying a channel that it
also thinks is free
A channel is used only if both sender and receiver conclude
that it is free
331
Impact of Directional Antennas
on MAC and Routing
332
Impact of Antennas on MAC

Wireless hosts traditionally use
single-mode antennas

Typically, the
single-mode = omni-directional

Recently, antennas with multiple modes (often, but
not necessarily, directional) have been develop

We will now focus on directional antennas with
multiple modes
333
IEEE 802.11

Implicitly assumes single mode antennas

Typically, omnidirectional antennas (though not
necessarily)
334
IEEE 802.11
Reserved area
RTS
A
B
C
D
E
F
CTS
335
Omni-Directional Antennas
Red nodes
Cannot
Communicate
presently
X
D
C
Y
336
Directional Antennas
Not possible
using Omni
X
D
C
Y
337
Question

How to exploit directional antennas in ad hoc
networks ?
Medium access control
Routing
338
Antenna Model
In Omni Mode:
 Nodes receive signals with gain Go
 While idle a node stays in omni mode
In Directional Mode:
 Capable of beamforming in specified direction
 Directional Gain Gd (Gd > Go)
 Directional mode has sidelobes
Symmetry: Transmit gain = Receive gain
339
Directional Communication
Received Power

(Transmit power) *(Tx Gain) * (Rx Gain)
Directional gain is higher
340
Potential Benefits of
Directional Antennas

Increase “range”, keeping transmit power constant

Reduce transmit power, keeping range comparable
with omni mode
Several proposal focus on this benefit
Assume that range of omni-directional and directional
transmission is equal
Directional transmissions at lower power
341
Caveats

Only most important features of the protocols
discussed here

Antenna characteristics assumed are often different
in different papers
342
Simple Tone Sense (STS) Protocol
[Yum1992IEEE Trans. Comm.]
343
STS Protocol
Based on busy tone signaling:
 Each host is assigned a tone (sinusoidal wave at a
certain frequency)
 Tone frequency unique in each host’s neighborhood
 When a host detects a packet destined to itself, it
transmit a tone
 If a host receive a tone on directional antenna A,it
assumes that some host in that direction is receiving
a packet
Cannot transmit using antenna A presently
OK to transmit using other antennas
344
STS Protocol

Tone duration used to encode information
Duration t1 implies transmitting node is busy
Duration t2 implies the transmitting node successfully
received a transmission from another node
345
Example
Node A cannot
Initiate a
transmission.
A
Tone t1
But B can send
to C
S
DATA
R
Because B does
not receive t1
B
C
346
STS Protocol
Issues:

Assigning tones to hosts

Assigning hosts to antennas: It is assumed that the
directions/angles can be chosen
distribute neighbor hosts evenly among the antennas
choose antenna angles such that adjacent antennas have
some minimum separation
347
Busy Tone Directional MAC
[Huang2002MILCOM]

Extends the busy tone (DBTMA) protocol originally
proposed by omni-directional antennas
[Deng98ICUPC]

Three channels
Data channel
Two Busy Tone channels
• Receive tone (BTr)
• Transmit tone (BTt)
348
DBTMA

Sender:
Sense BTr. If sensed busy, defer transmission.
If BTr idle, transmit RTS to receiver

Receiver
On receiving RTS, sense BTt.
If BTt idle, reply with a CTS, and transmit BTr until DATA is
completely received

Sender
On receiving CTS, transmit DATA and BTt both
349
DBTMA + Directional Antennas

DBTMA reduces reduction in throughput caused by
collisions by hidden terminals

Directional antennas can be used to transmit the
busy tones directionally
RTS/CTS, DATA, busy tones all may be sent directionally
Trade-offs similar to directional versus omni-directional
transmission of RTS/CTS
350
Another Directional MAC protocol
[Roychoudhury02mobicom]

Derived from IEEE 802.11 (similar to [Takai02mobihoc])

A node listens omni-directionally when idle
Sender transmits Directional-RTS (DRTS) towards receiver


RTS received in Omni mode (idle receiver in when idle)
Receiver sends Directional-CTS (DCTS)

DATA, ACK transmitted and received directionally

351
Directional MAC
RTS = Request-to-Send
X
RTS
A
B
C
D
E
F
Pretending a circular range for omni
352
Directional MAC
CTS = Clear-to-Send
X
CTS
A
B
C
D
E
F
353
Directional MAC
•DATA packet follows CTS. Successful data reception
acknowledged using ACK.
X
DATA
A
B
C
D
E
F
354
Directional MAC
X
ACK
A
B
C
D
E
F
355
Directional NAV (DNAV)
[Roychoudhury02mobicom]

Nodes overhearing RTS or CTS set up directional NAV (DNAV) for
that Direction of Arrival (DoA)
D
CTS
C
X
Y
Similar DNAV mechanism proposed in [Takai02mobihoc]
356
Directional NAV (DNAV)

Nodes overhearing RTS or CTS set up directional NAV (DNAV) for
that Direction of Arrival (DoA)
D
C
X
DNAV
Y
357
Directional NAV (DNAV)

New transmission initiated only if direction of transmission
does not overlap with DNAV,
i.e., if (θ > 0)
B
D
A
DNAV
θ
C
RTS
358
DMAC Example
C
E
D
B
B and C communicate
D and E cannot: D blocked with DNAV from C
D and A communicate
A
359
Issues with DMAC

Two types of Hidden Terminal Problems
Due to asymmetry in gain
Data
RTS
A
B
C
A is unaware of communication between B and C
A’s RTS may interfere with C’s reception of DATA
360
Issues with DMAC

Two types of Hidden Terminal Problems
Due to unheard RTS/CTS
D
B
A
C
Node A may now interfere at node C by transmitting
in C’s direction
361
Issues with DMAC
• Deafness
Z
RTS
A
B
DATA
RTS
Y
RTS
X
X does not know node A is busy.
X keeps transmitting RTSs to node A
Using omni antennas, X would be aware that A is
busy, and defer its own transmission
362
Issues with DMAC

Uses DO links, but not DD links
363
DMAC Tradeoffs

Benefits
Better Network
Connectivity
Spatial Reuse
• Disadvantages
– Hidden terminals
– Deafness
– No DD Links
364
Using Training Sequences
[Bellofiore2002IEEETrans.Ant.Prop]

Training packets used for DoA determination, after
RTS/CTS exchange omni-directionally
Sender
Receiver
RTS
RXTRN
CTS
DATA
TXTRN
ACK
365

Performance depends on the TXTRN and RXTRN
delays

If direction is known a priori, then these delays can
potentially be avoided
But mobility can change direction over time
366
Another Variation
[Nasipuri2000WCNC]

Similar to 802.11, but adapted for directional
antennas

Assumptions:
Antenna model: Several directional antennas which can all
be used simultaneously
Omni-directional reception is possible (by using all
directional antennas together)
Direction of arrival (DoA) can be determined when receiving
omni-directionally
Range of directional and omni transmissions are identical
367
Protocol Description


Sender sends omni-directional RTS
Receiver sends omni-directional CTS
Receiver also records direction of sender by determining the
antenna on which the RTS signal was received with highest
power level
Similarly, the sender, on receiving CTS, records the direction
of the receiver



All nodes overhearing RTS/CTS defer
transmissions
Sender then sends DATA directionally to the
receiver
Receiver sends directional ACK
368
Discussion

Protocol takes advantage of reduction in interference
due to directional transmission/reception of DATA

All neighbors of sender/receiver defer transmission
on receiving omni-directional RTS/CTS
 spatial reuse benefit not realized
369
Enhancing DMAC

Are improvements possible to make DMAC more
effective ?

Possible improvements:
Make Use of DD Links [Roychoudhury02MobiCom]
Overcome deafness [Roychoudhury03techrep]
370
Directional MAC: Summary

Directional antennas need adaptation of the MAC
protocols

MAC protocols show improvement in aggregate
throughput and delay
But not always

Performance dependent on topology
371
Routing with Directional Antennas
372
Routing Protocols

Many routing protocols for ad hoc networks rely on
broadcast messages
For instance, flood of route requests (RREQ)

Using omni antennas for broadcast will not discover
all possible links

Need to implement broadcast using directional
transmissions
373
Dynamic Source Routing [Johnson]

Sender floods RREQ through the network

Nodes forward RREQs after appending their names

Destination node receives RREQ and unicasts a
RREP back to sender node, using the route in which
RREQ traveled
374
Route Discovery in DSR
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
Represents a node that has received RREQ for D from S
375
Route Discovery in DSR
Y
Broadcast transmission
[S]
S
Z
E
F
B
C
M
J
A
L
G
H
K
I
D
N
Represents transmission of RREQ
[X,Y]
Represents list of identifiers appended to RREQ
376
Route Discovery in DSR
Y
Z
S
E
[S,E]
F
B
C
A
M
J
[S,C]
H
G
K
I
L
D
N
377
Route Discovery in DSR
Y
Z
S
E
F
B
[S,E,F]
C
M
J
A
L
G
H
I
[S,C,G] K
D
N
• Node C receives RREQ from G and H, but does not forward
it again, because node C has already forwarded RREQ once
378
Route Discovery in DSR
Y
Z
S
E
[S,E,F,J]
F
B
C
M
J
A
L
G
H
D
K
I
[S,C,G,K]
N
• Nodes J and K both broadcast RREQ to node D
379
Route Reply in DSR
Y
Z
S
E
RREP [S,E,F,J,D]
F
B
C
M
J
A
L
G
H
K
I
Represents RREP control message
D
N
380
DSR over Directional Antennas
[Roychoudhury03PWC,
Roychoudhury02UIUCTechrep]

RREQ broadcast by sweeping
To use DD links
381
Route Discovery in DSR
Y
Z
S
E
[S,E,F,J]
F
B
C
M
J
A
L
G
H
D
K
I
[S,C,G,K]
N
• Nodes J and K both broadcast RREQ to node D
382
Directional Routing
Broadcast by sweeping
Tradeoffs
Larger Tx Range
Few Hop Routes
Fewer Hop Routes
Low Data Latency
Small Beamwidth
More Sweeping
High Sweep Delay
High Overhead
383
Issues

Sub-optimal routes may be chosen if destination node
misses shortest request, while beamformed
F
D misses request from K
J
RREP
J
D
K RREQ

N
Optimize by having
destination wait before
replying
Broadcast storm: Using broadcasts, nodes receive
multiple copies of same packet
Use K antenna elements to
forward broadcast packet
384
Route Discovery in DSR
F
J
RREP
J
D
K
RREQ
N
D receives RREQ from J, and replies with RREP
D misses RREQ from K
385
Delayed RREP Optimization

Due to sweeping – earliest RREQ need not have
traversed shortest hop path.
RREQ packets sent to different neighbors at different points
of time

If destination replies to first arriving RREP, it might
miss shorter-path RREQ

Optimize by having DSR destination wait before
replying with RREP
386
Routing Overhead

Using omni broadcast, nodes receive multiple copies
of same packet - Redundant !!!
• Broadcast Storm Problem

Using directional Antennas – can do better ?
387
Routing Overhead
Use K antenna elements to forward broadcast packet
Footprint
of Tx
S (number of control packets)  (footprint of transmissions)
New measure for control
Overhead
=
S No. Data Packets
388
Mobility

Link lifetime increases using directional antennas.
Higher transmission range - link failures are less frequent

Nodes moving out of beam coverage in order of
packet-transmission-time
Low probability
389
Mobility

Antenna handoff
If no response to RTS, MAC layer uses N adjacent antenna
elements to transmit same packet
Route error avoided if communication re-established
[RoyChoudhury02UIUC Techrep]
390
Other Approaches to Routing
with Directional Antennas
[Nasipuri2000ICCCN]

Modified version of DSR

Transmit Route Request in the last known direction of
the receiver

If the source S perceives receiver R to have been in
direction d, then all nodes forward the route request
from S in direction d.
391
Example 1
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
392
Example 1
Y
Z
S
E
F
B
C
J
A
G
H
K
I
M
L
Route Reply
D
N
393
Example 2
Y
Z
S
E
F
B
C
M
J
A
L
G
H
K
I
D
N
D does
not
receive
394
RREQ
Limited Forwarding

Benefit: Limits the forwarding of the Route Request

Disadvantage: Effectively assumes that each node
has a sense of orientation
395
Routing with Directional Antennas: Conclusion

Directional antennas can improve routing
performance

But suitable protocol adaptations necessary
396
Directional Antennas: Conclusion

Directional antennas can potentially benefit

But also create difficulties in MAC and routing
protocol design
397
UDP on
Mobile Ad Hoc Networks
398
User Datagram Protocol (UDP)

UDP provides unreliable delivery

Studies comparing different routing protocols for
MANET typically measure UDP performance

Several performance metrics are often used
Routing overhead per data packet
Packet loss rate
Packet delivery delay
399
UDP Performance


Several relevant studies
[Broch98Mobicom,Das9ic3n,Johansson99Mobicom,
Das00Infocom,Jacquet00Inria]
Results comparing a specific pair of protocols do not
always agree, but some general (and intuitive)
conclusions can be drawn
Reactive protocols may yield lower routing overhead than
proactive protocols when communication density is low
Reactive protocols tend to loose more packets (assuming
than network layer drops packets if a route is not known)
Proactive protocols perform better with high mobility and
dense communication graph
400
UDP Performance

Many variables affect performance
Traffic characteristics
• one-to-many, many-to-one, many-to-many
• small bursts, large file transfers, real-time, non-real-time
Mobility characteristics
• low/high rate of movement
• do nodes tend to move in groups
Node capabilities
• transmission range (fixed, changeable)
• battery constraints
Performance metrics
• delay
• throughput
• latency
• routing overhead
Static or dynamic system characteristics (listed above)
401
UDP Performance

Difficult to identify a single scheme that will perform
well in all environments

Holy grail: Routing protocol that dynamically adapts
to all environments so as to optimize “performance”
Performance metrics may differ in different environments
402
TCP on
Mobile Ad Hoc Networks
403
Overview of
Transmission Control Protocol / Internet Protocol
(TCP/IP)
404
Internet Protocol (IP)

Packets may be delivered out-of-order

Packets may be lost

Packets may be duplicated
405
Transmission Control Protocol (TCP)

Reliable ordered delivery

Implements congestion avoidance and control

Reliability achieved by means of retransmissions if
necessary

End-to-end semantics
Acknowledgements sent to TCP sender confirm delivery of
data received by TCP receiver
Ack for data sent only after data has reached receiver
406
TCP Basics

Cumulative acknowledgements

An acknowledgement ack’s all contiguously received
data

TCP assigns byte sequence numbers
For simplicity, we will assign packet sequence
numbers
Also, we use slightly different syntax for acks than
normal TCP syntax
In our notation, ack i acknowledges receipt of packets


through packet i
407
Cumulative Acknowledgements

A new cumulative acknowledgement is generated
only on receipt of a new in-sequence packet
40
39
33
38
34
41
35
40
34
39
35
i
37
data
38
36
i
36
ack
37
408
Duplicate Acknowledgements

A dupack is generated whenever an
out-of-order segment arrives at the receiver
40
39
38
37
34
42
41
36
40
36
(Above example assumes delayed acks)
39
36
Dupack
On receipt of 38 409
Window Based Flow Control


Sliding window protocol
Window size minimum of
receiver’s advertised window - determined by available
buffer space at the receiver
congestion window - determined by the sender, based on
feedback from the network
Sender’s window
1 2 3 4 5 6 7 8 9 10 11 12 13
Acks received
Not transmitted
410
Window Based Flow Control
Sender’s window
1 2 3 4 5 6 7 8 9 10 11 12 13
Ack 5
1 2 3 4 5 6 7 8 9 10 11 12 13
Sender’s window
411
Window Based Flow Control

Congestion window size bounds the amount of data
that can be sent per round-trip time

Throughput <= W / RTT
412
Ideal Window Size

Ideal size = delay * bandwidth
delay-bandwidth product

What if window size < delay*bw ?
Inefficiency (wasted bandwidth)

What if > delay*bw ?
Queuing at intermediate routers
• increased RTT due to queuing delays
Potentially, packet loss
413
How does TCP detect a packet loss?

Retransmission timeout (RTO)

Duplicate acknowledgements
414
Detecting Packet Loss Using
Retransmission Timeout (RTO)

At any time, TCP sender sets retransmission timer for
only one packet

If acknowledgement for the timed packet is not
received before timer goes off, the packet is assumed
to be lost

RTO dynamically calculated
415
Retransmission Timeout (RTO) calculation

RTO = mean + 4 mean deviation
Standard deviation s : s 2 = average of (sample – mean)2
Mean deviation d = average of |sample – mean|
Mean deviation easier to calculate than standard deviation
Mean deviation is more conservative: d >= s
416
Exponential Backoff

Double RTO on each timeout
T1
T2 = 2 * T1
Timeout interval doubled
Packet
transmitted
Time-out occurs
before ack received,
packet retransmitted
417
Fast Retransmission

Timeouts can take too long
how to initiate retransmission sooner?

Fast retransmit
418
Detecting Packet Loss Using Dupacks:
Fast Retransmit Mechanism

Dupacks may be generated due to
packet loss, or
out-of-order packet delivery

TCP sender assumes that a packet loss has occurred
if it receives three dupacks consecutively
12 11 10 9 8 7
Receipt of packets 9, 10 and 11 will each generate
a dupack from the receiver. The sender, on getting
these dupacks, will retransmit packet 8.
419
Congestion Avoidance and Control

Slow Start: cwnd grows exponentially with time
during slow start

When cwnd reaches slow-start threshold, congestion
avoidance is performed

Congestion avoidance: cwnd increases linearly
with time during congestion avoidance
Rate of increase could be lower if sender does not always
have data to send
420
Congestion Window size
(segments)
14
Congestion
avoidance
12
10
8
Slow start
threshold
6Slow start
4
2
0
0
1
2
3
4
5
6
7
8
Time (round trips)
Example assumes that acks are not delayed
421
Congestion Control

On detecting a packet loss, TCP sender assumes
that network congestion has occurred

On detecting packet loss, TCP sender drastically
reduces the congestion window

Reducing congestion window reduces amount of data
that can be sent per RTT
422
Congestion Control -- Timeout

On a timeout, the congestion window is reduced to
the initial value of 1 MSS

The slow start threshold is set to half the window size
before packet loss
more precisely,
ssthresh = maximum of min(cwnd,receiver’s advertised
window)/2 and 2 MSS

Slow start is initiated
423
25
cwnd = 20
20
15
10
ssthresh = 10
ssthresh = 8
5
25
22
20
15
12
9
6
3
0
0
Congestion window (segments)
After timeout
Time (round trips)
424
Congestion Control - Fast retransmit

Fast retransmit occurs when multiple (>= 3) dupacks
come back

Fast recovery follows fast retransmit

Different from timeout : slow start follows timeout
timeout occurs when no more packets are getting across
fast retransmit occurs when a packet is lost, but latter
packets get through
ack clock is still there when fast retransmit occurs
no need to slow start
425
Fast Recovery

ssthresh =
min(cwnd, receiver’s advertised window)/2
(at least 2 MSS)



retransmit the missing segment (fast retransmit)
cwnd = ssthresh + number of dupacks
when a new ack comes: cwnd = ssthreh
enter congestion avoidance
Congestion window cut into half
426
Window size (segments)
After fast recovery
10
Receiver’s advertised window
8
6
4
2
0
0
2
4
6
8
10 12 14
Time (round trips)
After fast retransmit and fast recovery window size is
reduced in half.
427
TCP Reno




Slow-start
Congestion avoidance
Fast retransmit
Fast recovery
428
TCP Performance
in
Mobile Ad Hoc Networks
429
Performance of TCP
Several factors affect TCP performance in MANET:

Wireless transmission errors

Multi-hop routes on shared wireless medium
For instance, adjacent hops typically cannot transmit
simultaneously

Route failures due to mobility
430
Random Errors

If number of errors is small, they may be corrected by
an error correcting code

Excessive bit errors result in a packet being
discarded, possibly before it reaches the transport
layer
431
Random Errors May Cause Fast Retransmit
40
39
38
37
34
36
Example assumes delayed ack - every other packet ack’d
432
Random Errors May Cause Fast Retransmit
41
40
34
39
38
36
Example assumes delayed ack - every other packet ack’d
433
Random Errors May Cause Fast Retransmit
42
41
40
36
39
36
dupack
Duplicate acks are not delayed
434
Random Errors May Cause Fast Retransmit
43
42
36
41
40
36
36
Duplicate acks
435
Random Errors May Cause Fast Retransmit
44
43
42
36
41
36
36
3 duplicate acks trigger
fast retransmit at sender
436
Random Errors May Cause Fast Retransmit

Fast retransmit results in
retransmission of lost packet
reduction in congestion window

Reducing congestion window in response to errors is
unnecessary

Reduction in congestion window reduces the
throughput
437
Sometimes Congestion Response May be
Appropriate in Response to Errors

On a CDMA channel, errors occur due to interference
from other user, and due to noise [Karn99pilc]
Interference due to other users is an indication of
congestion. If such interference causes transmission errors,
it is appropriate to reduce congestion window
If noise causes errors, it is not appropriate to reduce window

When a channel is in a bad state for a long duration,
it might be better to let TCP backoff, so that it does
not unnecessarily attempt retransmissions while the
channel remains in the bad state
[Padmanabhan99pilc]
438
Impact of Random Errors [Vaidya99]
1600000
1200000
800000
bits/sec
400000
0
16384 32768 65536 131072
1/error rate
(in bytes)
Exponential error model
2 Mbps wireless full duplex link
No congestion losses
439
Burst Errors May Cause Timeouts




If wireless link remains unavailable for extended
duration, a window worth of data may be lost
driving through a tunnel
passing a truck
Timeout results in slow start
Slow start reduces congestion window to 1 MSS,
reducing throughput
Reduction in window in response to errors
unnecessary
440
Random Errors May Also Cause Timeout

Multiple packet losses in a window can result in
timeout when using TCP-Reno (and to a lesser extent
when using SACK)
441
Impact of Transmission Errors

TCP cannot distinguish between packet losses due to
congestion and transmission errors

Unnecessarily reduces congestion window

Throughput suffers
442
This Tutorial

This tutorial does not consider techniques to improve
TCP performance in presence of transmission errors

Please refer to the Tutorial on TCP for Wireless and
Mobile Hosts presented by Vaidya at MobiCom 1999,
Seattle
The tutorial slides are presently available from
http://www.cs.tamu.edu/faculty/vaidya/ (follow the link to
Seminars)

[Montenegro00-RFC2757] discusses related issues
443
This Tutorial

This tutorial considers impact of multi-hop routes and
route failures due to mobility
444
Mobile Ad Hoc Networks

May need to traverse multiple links to reach a
destination
445
Mobile Ad Hoc Networks

Mobility causes route changes
446
Throughput over Multi-Hop Wireless Paths
[Gerla99]

Connections over multiple hops are at a
disadvantage compared to shorter connections,
because they have to contend for wireless access at
each hop
447
Impact of Multi-Hop Wireless Paths
[Holland99]
1600
1400
1200
1000
800
600
400
200
0
TCP
Throughtput
(Kbps)
1 2 3 4 5 6 7 8 9 10
Number of hops
TCP Throughput using 2 Mbps 802.11 MAC
448
Throughput Degradations with
Increasing Number of Hops

Packet transmission can occur on at most one hop
among three consecutive hops
Increasing the number of hops from 1 to 2, 3 results in
increased delay, and decreased throughput

Increasing number of hops beyond 3 allows
simultaneous transmissions on more than one link,
however, degradation continues due to contention
between TCP Data and Acks traveling in opposite
directions

When number of hops is large enough, the
throughput stabilizes due to effective pipelining
449
Ideal Throughput

f(i) = fraction of time for which shortest path length
between sender and destination is I

T(i) = Throughput when path length is I
From previous figure

Ideal throughput = S f(i) * T(i)
450
Impact of Mobility
TCP Throughput
2 m/s
10 m/s
Ideal throughput (Kbps)
451
Impact of Mobility
20 m/s
30 m/s
Ideal throughput
452
Throughput generally degrades with increasing
speed …
Ideal
Average
Throughput
Over
50 runs
Actual
Speed (m/s)
453
But not always …
30 m/s
20 m/s
Actual
throughput
Mobility pattern #
454
Why Does Throughput Degrade?
mobility causes
link breakage,
resulting in route
failure
Route is
repaired
TCP sender times out.
Starts sending packets again
No throughput
No throughput
despite route repair
TCP data and acks
en route discarded
455
Why Does Throughput Degrade?
mobility causes
link breakage,
resulting in route
failure
TCP sender
times out.
Backs off timer.
Route is
repaired
TCP sender
times out.
Resumes
sending
No throughput
No throughput
despite route repair
Larger route repair delays
especially harmful
TCP data and acks
en route discarded
456
Why Does Throughput Improve?
Low Speed Scenario
C
B
D
C
D
B
A
C
D
B
A
A
1.5 second route failure
Route from A to D is broken for ~1.5 second.
When TCP sender times after 1 second, route still broken.
TCP times out after another 2 seconds, and only then resumes.
457
Why Does Throughput Improve?
Higher (double) Speed Scenario
C
B
D
C
D
B
A
C
D
B
A
A
0.75 second route failure
Route from A to D is broken for ~ 0.75 second.
When TCP sender times after 1 second, route is repaired.
458
Why Does Throughput Improve?
General Principle

The previous two slides show a plausible cause for
improved throughput

TCP timeout interval somewhat (not entirely)
independent of speed

Network state at higher speed, when timeout occurs,
may be more favorable than at lower speed

Network state
Link/route status
Route caches
Congestion
459
How to Improve Throughput
(Bring Closer to Ideal)

Network feedback

Inform TCP of route failure by explicit message

Let TCP know when route is repaired
Probing
Explicit notification

Reduces repeated TCP timeouts and backoff
460
Performance Improvement
With feedback
Actual throughput
Without network
feedback
Ideal throughput
2 m/s speed
461
Performance Improvement
With feedback
Actual throughput
Without network
feedback
Ideal throughput
30 m/s speed
462
throughput as a fraction of
ideal
Performance with Explicit Notification
[Holland99]
1
0.8
Base TCP
0.6
With explicit
notification
0.4
0.2
0
2
10
20
30
mean speed (m/s)
463
Issues
Network Feedback

Network knows best (why packets are lost)
+ Network feedback beneficial
- Need to modify transport & network layer to
receive/send feedback

Need mechanisms for information exchange between
layers

[Holland99] discusses alternatives for providing
feedback (when routes break and repair)
[Chandran98] also presents a feedback scheme
464
Impact of Caching

Route caching has been suggested as a mechanism
to reduce route discovery overhead [Broch98]

Each node may cache one or more routes to a given
destination

When a route from S to D is detected as broken,
node S may:
Use another cached route from local cache, or
Obtain a new route using cached route at another node
465
To Cache or Not to Cache
Average speed (m/s)
466
Why Performance Degrades With Caching

When a route is broken, route discovery returns a
cached route from local cache or from a nearby node

After a time-out, TCP sender transmits a packet on
the new route.
However, the cached route has also broken after it
was cached
timeout due
to route failure


timeout, cached timeout, second cached
route is broken
route also broken
Another route discovery, and TCP time-out interval
Process repeats until a good route is found
467
Issues
To Cache or Not to Cache

Caching can result in faster route “repair”

Faster does not necessarily mean correct

If incorrect repairs occur often enough, caching
performs poorly

Need mechanisms for determining when cached
routes are stale
468
Caching and TCP performance

Caching can reduce overhead of route discovery
even if cache accuracy is not very high

But if cache accuracy is not high enough, gains in
routing overhead may be offset by loss of TCP
performance due to multiple time-outs
469
TCP Performance
Two factors result in degraded throughput in presence
of mobility:

Loss of throughput that occurs while waiting for TCP
sender to timeout (as seen earlier)
This factor can be mitigated by using explicit notifications
and better route caching mechanisms

Poor choice of congestion window and RTO values
after a new route has been found
How to choose cwnd and RTO after a route change?
470
Issues
Window Size After Route Repair

Same as before route break: may be too optimistic

Same as startup: may be too conservative

Better be conservative than overly optimistic
Reset window to small value after route repair
Let TCP figure out the suitable window size
Impact low on paths with small delay-bw product
471
Issues
RTO After Route Repair

Same as before route break

Same as TCP start-up (6 second)

Another plausible approach: new RTO = function of old RTO, old
route length, and new route length
If new route long, this RTO may be too small, leading to timeouts
May be too large
May result in slow response to next packet loss
Example: new RTO = old RTO * new route length / old route length
Not evaluated yet
Pitfall: RTT is not just a function of route length
472
Out-of-Order Packet Delivery



Out-of-order (OOO) delivery may occur due to:
Route changes
Link layer retransmissions schemes that deliver OOO
Significantly OOO delivery confuses TCP, triggering
fast retransmit
Potential solutions:
Deterministically prefer one route over others, even if
multiple routes are known
Reduce OOO delivery by re-ordering received packets
can result in unnecessary delay in presence of packet
loss
Turn off fast retransmit
can result in poor performance in presence of congestion
•
•
473
Impact of Acknowledgements

TCP Acks (and link layer acks) share the wireless
bandwidth with TCP data packets

Data and Acks travel in opposite directions
In addition to bandwidth usage, acks require additional
receive-send turnarounds, which also incur time penalty
To reduce frequency of send-receive turnaround and
contention between acks and data
474
Impact of Acks: Mitigation [Balakrishnan97]

Piggybacking link layer acks with data

Sending fewer TCP acks - ack every d-th packet (d
may be chosen dynamically)
• but need to use rate control at sender to reduce
burstiness (for large d)

Ack filtering - Gateway may drop an older ack in the
queue, if a new ack arrives
reduces number of acks that need to be delivered to the
sender
475
Security Issues
476
Caveat

Much of security-related stuff is mostly beyond my
expertise

So coverage of this topic is very limited
477
Security Issues in Mobile Ad Hoc Networks

Not much work in this area as yet

Many of the security issues are same as those in
traditional wired networks and cellular wireless

What’s new ?
478
What’s New ?

Wireless medium is easy to snoop on

Due to ad hoc connectivity and mobility, it is hard to
guarantee access to any particular node (for
instance, to obtain a secret key)

Easier for trouble-makers to insert themselves into a
mobile ad hoc network (as compared to a wired
network)
479
Resurrecting Duckling [Stajano99]

Battery exhaustion threat: A malicious node may
interact with a mobile node often with the goal of
draining the mobile node’s battery

Authenticity: Who can a node talk to safely?
Resurrecting duckling: Analogy based on a duckling and its
mother. Apparently, a duckling assumes that the first object it
hears is the mother
A mobile device will trust first device which sends a secret
key
480
Secure Routing [Zhou99]

Attackers may inject erroneous routing information

By doing so, an attacker may be able to divert
network traffic, or make routing inefficient

[Zhou] suggests use of digital signatures to protect
routing information and data both

Such schemes need a Certification Authority to
manage the private-public keys
481
Secure Routing

Establishing a Certification Authority (CA) difficult in a
mobile ad hoc network, since the authority may not
be reachable from all nodes at all times

[Zhou] suggests distributing the CA function over
multiple nodes
482
MANET Authentication Architecture
[Jacobs99ietf-id]

Digital signatures to authenticate a message

Key distribution via certificates

Need access to a certification authority

[Jacobs99ietf-id] specifies message formats to be
used to carry signature, etc.
483
Techniques for Intrusion-Resistant Ad Hoc
Routing Algorithms (TIARA) [Ramanujan00Milcom]

Flow disruption attack: Intruder (or compromised)
node T may delay/drop/corrupt all data passing
through, but leave all routing traffic unmodified
B
C
A
D
T
intruder
484
Techniques for Intrusion-Resistant Ad Hoc
Routing Algorithms (TIARA) [Ramanujan00Milcom]

Resource Depletion Attack: Intruders may send data
with the objective of congesting a network or
depleting batteries
U
B
intruder
C
A
D
T
Bogus traffic
intruder
485
Intrusion Detection [Zhang00Mobicom]

Detection of abnormal routing table updates
Uses “training” data to determine characteristics of normal
routing table updates (such as rate of change of routing info)
Efficacy of this approach is not evaluated, and is debatable

Similar abnormal behavior may be detected at other
protocol layers
For instance, at the MAC layer, normal behavior may be
characterized for access patterns by various hosts
Abnormal behavior may indicate intrusion

Solutions proposed in [Zhang00Mobicom] are
preliminary, not enough detail provided
486
Preventing Traffic Analysis
[Jiang00iaas,Jiang00tech]

Even with encryption, an eavesdropper may be able
to identify the traffic pattern in the network

Traffic patterns can give away information about the
mode of operation
Attack versus retreat

Traffic analysis can be prevented by presenting
“constant” traffic pattern independent of the
underlying operational mode
May need insertion of dummy traffic to achieve this
487
Packet Purse Model [Byttayn00MobiHoc]

Cost-based approach for motivating collaboration
between mobile nodes

The packet purse model assigns a cost to each
packet transfer
Link-level recipient of a packet pays the link-level sender for
the service
Virtual money (“beans”) used for this purpose

Security issues:
How to ensure that some node does not sale the same
packet to too many people to make money ?
How to ensure that each receiver indeed has money to pay488
for service?
MAC Layer Misbehavior
489
Selfish Misbehavior to Improve Performance
Access Point

Wireless
channel
A
Nodes are required to
follow Medium Access
Control (MAC) rules
B
Misbehaving nodes may violate MAC rules
490
Backoff Example

Choose backoff value B in range [0,CW]
CW is the Contention Window

Count down backoff by 1 every idle slot
B1=15
S1
B1=0
B1=20
Transmit
wait
wait
Transmit
CW=31
S2
B2=25
B2=10
B2=10
491
Data Transmission

Reserve channel with RTS/CTS exchange
RTS
A
CTS
S
R
B
B=10
Sender
S
Receiver
R
492
Possible Misbehavior

Backoff from biased distribution
Example: Always select a small backoff value
B1 = 1
B1 = 1
Misbehaving
node
Transmit
Transmit
Well-behaved
node
wait
wait
B2 = 20
B2 = 19
493
Goals [Kyasanur03dsn]

Diagnose node misbehavior
 Catch misbehaving nodes

Discourage misbehavior
 Punish misbehaving nodes
494
MAC Selfishness: Game-Theoretic Approach

MacKenzie addresses selfish misbehavior in Aloha
networks
Nodes may use higher access probabilities

Solution uses game theoretic approach
Assumes there is some cost for transmitting
Nodes independently adjust access prob.
Under some assumptions network reaches a fair equilibrium
495
MAC: Selfishness

[Konorski01, Konorski02] discuss selfish misbehavior
in 802.11 networks

Game theory used to analyze solution
Nodes use a black-burst to resolve contention
Winner is not the largest burst, but node with burst within 
slots of largest burst
496
Game theory - Discussion

Protocols resilient to misbehavior can be developed
Do not need explicit misbehavior detection

Solutions assume perfect knowledge
No guarantees with imperfect information

Performance at equilibrium may be poor
497
Alternative Approach

Use payment schemes, charging per packet

Misbehaving node can achieve lower delay (e.g., by
sending packet bursts)
 Average delay is less with same cost
Per-packet payment schemes not sufficient
(need to factor delay – harder)
498
Another Approach

Receivers detect sender misbehavior
Assume receivers are well-behaved (can be relaxed)
Access Point
A

Access Point is trusted.
When AP transmits, it is wellbehaved

When AP receives, it can
monitor sender behavior
Wireless
channel
499
Issues

Receiver does not know exact backoff value
chosen by sender
 Sender chooses random backoff
 Hard to distinguish between maliciously chosen small values
and a legitimate random sequence

Wireless channel introduces uncertainties
 Channel status seen by sender and receiver may be different
500
Potential Solution:
Use long-term statistics

Observe backoffs chosen by sender over multiple packets

Backoff values not from expected distribution 
Misbehavior
Selecting right observation interval difficult
501
A Simpler Approach

Remove the non-determinism
502
A Simpler Approach

Receiver provides backoff values to sender
Receiver specified backoff for next packet in ACK for current
packet

Modification does not significantly change 802.11
behavior
Backoffs of different nodes still independent
Uncertainty of sender’s backoff eliminated
503
Modifications to 802.11
B
Sender
S
Receiver
R
• R provides backoff B to S in ACK
B selected from [0,CWmin]
• S uses B for backoff
504
Protocol steps
Step 1: For each transmission:
 Detect deviations: Decide if sender backed off for less than

required number of slots
Penalize deviations: Penalty is added, if the sender appears to
have deviated
Goal: Identify and penalize suspected misbehavior
 Reacting to individual transmission makes it harder to adapt to
the protocol
505
Protocol steps
Step 2: Based on last W transmissions:
 Diagnose misbehavior: Identify misbehaving nodes
Goal: Identify misbehaving nodes with high probability
 Reduce impact of channel uncertainties
 Filter out misbehaving nodes from well-behaved nodes
506
Detecting deviations
Backoff
Sender S
Receiver R
Bobsr

Receiver counts number of idle slots Bobsr
Condition for detecting deviations: Bobsr < a B
(0 < a <= 1)
507
Penalizing Misbehavior
Actual backoff < B
Sender
S
Receiver
R
Bobsr
 When Bobsr < a B, penalty P added

P proportional to a B– Bobsr
 Total backoff assigned = B + P
508
Penalty Scheme issues

Misbehaving sender has two options
Ignore assigned penalty  Easier to detect
Follow assigned penalty  No throughput gain

With penalty, sender has to misbehave more for same
throughput gain
509
Diagnosing Misbehavior

Total deviation for last W packets used
Deviation per packet is B – Bobsr

If total deviation > THRESH then sender is designated as
misbehaving

Higher layers / administrator can be informed of
misbehavior
510
MANET
Implementation Issues
511
Existing Implementations

Several implementations apparently exist (see IETF
MANET web site)

Only a few available publicly

Most implementations focus on unicast routing
512
CMU Implementation [Maltz99]
User space
TCP/UDP
Kernel space
DSR option processing (RREQ, RREP,…)
Route cache
Send
buffer
rexmit
buffer
IP
Route table
DSR Output
dsr_xmit
Kernel space
Physical
devices
WaveLan-I
CDPD
513
CMU Implementation: Lessons Learned

Multi-level priority queues helpful: Give higher priority
to routing control packets, and lower for data

If retransmission is implemented above the link layer,
it must be adaptive to accommodate congestion
Since Wavelan-I MAC does not provide retransmissions,
DSR performs retransmits itself
DSR per-hop ack needs to contend for wireless medium
Time to get the ack (RTT) is dependent on congestion
TCP-like RTT estimation and RTO used for triggering
retransmits by DSR on each hop
This is not very relevant when using IEEE 802.11 where the
ack is sent immediately after data reception
514
CMU Implementation: Lessons Learned

“Wireless propagation is not what you would expect”
[Maltz99]
Straight flat areas with line-of-sight connectivity had worst
error rates

“Bystanders will think you are nuts” [Maltz99]
If you are planning experimental studies in the streets, it
may be useful to let police and security guards know in
advance what you are up to
515
BBN Implementation [Ramanathan00Wcnc]

Density and Asymmetric-Adaptive Wireless Network
(DAWN)
Quote from [Ramanathan00Wcnc]: DAWN is a “subnet” or
“link” level system from IP’s viewpoint and runs “below” IP
Nokia IP Stack
DAWN IP Gateway
DAWN
Protocols
=
Topology
control
Scalable
Link State
Routing
Elastic
Virtual
Circuits
Qos Based Forwarding
Nokia MAC
Utilicom 2050 Radio
516
DAWN Features

Topology control by transmit power control
To avoid topologies that are too sparse or too dense
To extend battery life

Scalable link state routing: Link state updates with
small TTL (time-to-live) sent more often, than those
with greater TTL
As a packet gets closer to the destination, more accurate
info is used for next hop determination

Elastic Virtual Circuits (VC):
Label switching through the DAWN nodes (label = VC id)
Path repaired transparent to the endpoints when hosts along
517
the path move away
Implementation Issues:
Where to Implement Ad Hoc Routing

Link layer

Network layer

Application layer
518
Implementation Issues:
Security

How can I trust you to forward my packets without
tampering?
Need to be able to detect tampering

How do I know you are what you claim to be ?
Authentication issues
Hard to guarantee access to a certification authority
519
Implementation Issues

Can we make any guarantees on performance?
When using a non-licensed band, difficult to provide hard
guarantees, since others may be using the same band

Must use an licensed channel to attempt to make any
guarantees
520
Implementation Issues

Only some issues have been addresses in existing
implementations

Security issues often ignored

Address assignment issue also has not received
sufficient attention
521
Integrating MANET with the Internet [Broch99]

Mobile IP + MANET routing

At least one node in a MANET should act as a
gateway to the rest of the world

Such nodes may be used as foreign agents for
Mobile IP

IP packets would be delivered to the foreign agent of
a MANET node using Mobile IP. Then, MANET
routing will route the packet from the foreign agent to
the mobile host.
522
Related Standards Activities
523
Internet Engineering Task Force (IETF)
Activities

IETF manet (Mobile Ad-hoc Networks) working
group
http://www.ietf.org/html.charters/manet-charter.html

IETF mobileip (IP Routing for Wireless/Mobile
Hosts) working group
http://www.ietf.org/html.charters/mobileip-charter.html
524
Internet Engineering Task Force (IETF)
Activities

IETF pilc (Performance Implications of Link
Characteristics) working group
http://www.ietf.org/html.charters/pilc-charter.html
http://pilc.grc.nasa.gov
Refer [RFC2757] for an overview of related work
525
Related Standards Activities

BlueTooth
http://www.bluetooth.com

HomeRF [Lansford00ieee]
http://www.homerf.org

IEEE 802.11
http://grouper.ieee.org/groups/802/11/

Hiperlan/2
http://www.etsi.org/technicalactiv/hiperlan2.htm
526
Bluetooth
[Haartsen98,Bhagawat00Tutorial]

Features: Cheaper, smaller, low power, ubiquitous,
unlicensed frequency band

Spec version 1.0B released December 1999
(1000+ pages)

Promoter group consisting of 9
Ericsson, IBM, Intel, Nokia, Toshiba, 3Com, Lucent,
Microsoft, Motorola

1800+ adopters
527
Bluetooth: Link Types


Designed to support multimedia applications that mix
voice and data
Synchronous Connection-Oriented (SCO) link
Symmetrical, circuit-switched, point-to-point connections
Suitable for voice
Two consecutive slots (forward and return slots) reserved at
fixed intervals

Asynchronous Connectionless (ACL) link
Symmetrical or asymmetric, packet-switched, point-tomultipoint
Suitable for bursty data
Master units use a polling scheme to control ACL
connections
528
Bluetooth: Piconet





A channel is characterized by a frequency-hopping
pattern
Two or more terminals sharing a channel form a
piconet
1 Mbps per Piconet
One terminal in a piconet acts as a master and up to
7 slaves
Other terminals are slaves
Polling scheme: A slave may send in a slave-tomaster slot when it has been addressed by its MAC
address in the previous master-to-slave slot
529
Inter-Piconet Communication

A slave can belong to two different piconets, but not
at the same time

A slave can leave its current piconet (after informing
its current master the duration of the leave) and join
another piconet

A maser of one piconet can also join another piconet
temporarily as a slave
530
Bluetooth: Scatternet

Several piconets may exist in the same area (such
that units in different piconets are in each other’s
range)

Each piconet uses a different channel and gets 1
Mbps for the piconet
Since two independently chosen hopping patterns may
select same hop simultaneously with non-zero probability,
some collisions between piconets are possible, reducing
effective throughput

A group of piconets is called a scatternet
531
Routing

Ad hoc routing protocols needed to route between
multiple piconets

Existing protocols may need to be adapted for
Bluetooth [Bhagwat99Momuc]
For instance, not all nodes within transmission range of
node X will hear node X
Only nodes which belong to node X’s current piconet can
hear the transmission from X
Flooding-based schemes need to take this limitation into
account
•
532
Open Issues
in
Mobile Ad Hoc Networking
533
Open Problems

Issues other than routing have received much less
attention so far
Other interesting problems:






Address assignment problem
MAC protocols
Improving interaction between protocol layers
Distributed algorithms for MANET
QoS issues
Applications for MANET
534
Related Research Areas

Algorithms for dynamic networks (e.g., [Afek89])

Sensor networks [DARPA-SensIT]
Ad hoc network of sensors
Addressing based on data (or function) instead of name
• “send this packet to a temperature sensor”
535
References

Please see attached listing for the references cited in
the tutorial
536
Thank you !!
For more information, send e-mail to
Nitin Vaidya at
nhv@uiuc.edu
© 2004 Nitin Vaidya
537